scripting

apps/web/scripts/fine-tune-model.py

@@ -0,0 +1,537 @@
+#!/usr/bin/env python3
+"""
+fine-tune-model.py
+
+Fine-tunes the PlantVillage MobileNetV2 model on a custom 95-class dataset
+(93 diseases + healthy + unknown).
+
+Pipeline:
+  1. Load `best_mnv2_pv_original.keras` (MobileNetV2 backbone + 38-class head)
+  2. Replace the 38-class head with 95 classes (order matches diseases.json + healthy + unknown)
+  3. Freeze backbone, train only the new classification head
+  4. Unfreeze the last ~20 layers, fine-tune at lower learning rate
+  5. Export to TF.js GraphModel format
+  6. Export to .keras for future retraining
+
+Usage: .tfjs-venv/bin/python scripts/fine-tune-model.py
+"""
+
+import json
+import os
+import sys
+import shutil
+from pathlib import Path
+
+os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2"  # Suppress TF info/warnings
+
+import numpy as np
+import tensorflow as tf
+import keras
+from keras import layers, optimizers, regularizers
+
+# ─── Constants ───────────────────────────────────────────────────────────────
+
+PROJECT_ROOT = Path(__file__).resolve().parent.parent
+MODEL_PATH = (
+    PROJECT_ROOT
+    / "public"
+    / "models"
+    / "plant-disease-classifier"
+    / "best_mnv2_pv_original.keras"
+)
+DISEASES_JSON = PROJECT_ROOT / "src" / "data" / "diseases.json"
+DATASET_DIR = PROJECT_ROOT / "data" / "dataset"
+OUTPUT_DIR = PROJECT_ROOT / "public" / "models" / "plant-disease-classifier"
+TFJS_OUTPUT = OUTPUT_DIR / "tfjs_finetuned"
+
+IMG_SIZE = 160  # Model input size
+BATCH_SIZE = 32
+EPOCHS_HEAD = 15  # Train just the new head
+EPOCHS_FINETUNE = 10  # Unfreeze and fine-tune
+LEARNING_RATE_HEAD = 1e-3
+LEARNING_RATE_FINETUNE = 1e-5
+VALIDATION_SPLIT = 0.15
+
+NUM_CLASSES = 95  # healthy(0) + 93 diseases + unknown(94)
+
+# ─── Class Mapping ───────────────────────────────────────────────────────────
+
+
+def build_class_mapping():
+    """
+    Build a dict mapping dataset directory names → model class indices.
+    Matches the ordering in labels.ts / diseases.json.
+
+    Index 0  = "healthy"
+    Index 1-93 = disease IDs (in diseases.json order)
+    Index 94 = "unknown" (no images — skip during training)
+    """
+    with open(DISEASES_JSON) as f:
+        diseases = json.load(f)
+
+    mapping = {"healthy": 0}
+    for i, disease in enumerate(diseases):
+        mapping[disease["id"]] = i + 1  # Index 1-93
+    mapping["unknown"] = 94  # Not trained, but reserved
+
+    # Reverse mapping for predictions
+    index_to_class = {v: k for k, v in mapping.items()}
+
+    return mapping, index_to_class
+
+
+def verify_dataset(mapping):
+    """Find which classes have images and how many."""
+    available = {}
+    total = 0
+
+    for class_id, class_idx in mapping.items():
+        class_dir = DATASET_DIR / class_id
+        if not class_dir.exists():
+            continue
+
+        image_paths = sorted(class_dir.glob("*"))
+        image_paths = [
+            p
+            for p in image_paths
+            if p.suffix.lower() in (".jpg", ".jpeg", ".png", ".webp")
+        ]
+
+        if image_paths:
+            available[class_id] = {"index": class_idx, "count": len(image_paths)}
+            total += len(image_paths)
+
+    return available, total
+
+
+def print_dataset_summary(available, total):
+    """Print a summary of what's available."""
+    print(f"\n{'─' * 60}")
+    print("DATASET SUMMARY")
+    print(f"{'─' * 60}")
+    print(f"  Total images: {total}")
+    print(f"  Classes found: {len(available)} / {len(build_class_mapping()[0])}")
+    print(
+        f"  Missing classes with no images: {len(build_class_mapping()[0]) - len(available)}"
+    )
+
+    # Count images per class
+    counts = [(v["index"], k, v["count"]) for k, v in available.items()]
+    counts.sort(key=lambda x: x[1])
+
+    print("\n  Images per class:")
+    for idx, class_id, count in counts:
+        label = f"  {idx:3d}. {class_id:<35s} {count:>4d} images"
+        if class_id == "healthy":
+            label += " ← 2× target"
+        print(label)
+
+    # Stats
+    class_counts = [v["count"] for v in available.values()]
+    if class_counts:
+        print(
+            f"\n  Min: {min(class_counts)}  Max: {max(class_counts)}  Avg: {sum(class_counts) / len(class_counts):.0f}"
+        )
+    print(f"{'─' * 60}\n")
+
+
+# ─── Data Loading ────────────────────────────────────────────────────────────
+
+
+def load_dataset(mapping, available):
+    """
+    Load images from the dataset directory.
+    Returns train/validation datasets with augmentation.
+    """
+    # Build file paths and labels
+    file_paths = []
+    labels = []
+
+    for class_id, info in available.items():
+        class_dir = DATASET_DIR / class_id
+        images = sorted(class_dir.glob("*"))
+        images = [
+            p for p in images if p.suffix.lower() in (".jpg", ".jpeg", ".png", ".webp")
+        ]
+
+        for img_path in images:
+            file_paths.append(str(img_path))
+            labels.append(info["index"])
+
+    file_paths = np.array(file_paths)
+    labels = np.array(labels)
+
+    # Shuffle
+    indices = np.random.RandomState(42).permutation(len(file_paths))
+    file_paths = file_paths[indices]
+    labels = labels[indices]
+
+    # Split train/validation
+    split = int(len(file_paths) * (1 - VALIDATION_SPLIT))
+    train_paths, val_paths = file_paths[:split], file_paths[split:]
+    train_labels, val_labels = labels[:split], labels[split:]
+
+    print(f"  Train: {len(train_paths)} images")
+    print(f"  Val:   {len(val_paths)} images")
+
+    # Parse function
+    def parse_image(image_path, label):
+        img = tf.io.read_file(image_path)
+        img = tf.image.decode_image(img, channels=3, expand_animations=False)
+        img = tf.image.resize(img, [IMG_SIZE, IMG_SIZE])
+        img = tf.cast(img, tf.float32) / 255.0
+        # ImageNet normalization (matching training-time preprocessing)
+        mean = tf.constant([0.485, 0.456, 0.406])
+        std = tf.constant([0.229, 0.224, 0.225])
+        img = (img - mean) / std
+        return img, label
+
+    def augment(image, label):
+        """Data augmentation for training set."""
+        # Random horizontal flip
+        image = tf.image.random_flip_left_right(image)
+        # Random rotation (±20°)
+        image = tf.image.random_flip_up_down(image)
+        # Random brightness
+        image = tf.image.random_brightness(image, 0.15)
+        # Random contrast
+        image = tf.image.random_contrast(image, 0.8, 1.2)
+        # Random saturation
+        image = tf.image.random_saturation(image, 0.8, 1.2)
+        # Random hue
+        image = tf.image.random_hue(image, 0.05)
+        # Random crop (after slightly scaling up)
+        image = tf.image.resize_with_crop_or_pad(image, IMG_SIZE + 12, IMG_SIZE + 12)
+        image = tf.image.resize(image, [IMG_SIZE, IMG_SIZE])
+        # Clip to valid range after augmentations
+        image = tf.clip_by_value(image, -2.5, 2.5)
+        return image, label
+
+    # Create tf.data datasets
+    train_ds = tf.data.Dataset.from_tensor_slices((train_paths, train_labels))
+    train_ds = train_ds.map(parse_image, num_parallel_calls=tf.data.AUTOTUNE)
+    train_ds = train_ds.map(augment, num_parallel_calls=tf.data.AUTOTUNE)
+    train_ds = train_ds.shuffle(1000).batch(BATCH_SIZE).prefetch(tf.data.AUTOTUNE)
+
+    val_ds = tf.data.Dataset.from_tensor_slices((val_paths, val_labels))
+    val_ds = val_ds.map(parse_image, num_parallel_calls=tf.data.AUTOTUNE)
+    val_ds = val_ds.batch(BATCH_SIZE).prefetch(tf.data.AUTOTUNE)
+
+    return train_ds, val_ds
+
+
+# ─── Model Building ──────────────────────────────────────────────────────────
+
+
+def build_model():
+    """
+    Load the PlantVillage model and replace the classification head
+    with a 95-class output.
+    """
+    print(f"\nLoading base model from: {MODEL_PATH}")
+    if not MODEL_PATH.exists():
+        print(f"ERROR: Model not found at {MODEL_PATH}")
+        sys.exit(1)
+
+    base_model = keras.models.load_model(str(MODEL_PATH))
+    print(f"  Base model loaded: {type(base_model).__name__}")
+    print(f"  Input shape: {base_model.input_shape}")
+    print(f"  Output shape: {base_model.output_shape}")
+
+    # Extract backbone — everything up to the GlobalAveragePooling2D
+    # The model structure is:
+    #   input_layer_2 → mobilenetv2_1.00_160 → global_average_pooling2d → dropout → dense(38)
+    backbone_output = base_model.get_layer("global_average_pooling2d").output
+    print("  Using backbone output: global_average_pooling2d")
+
+    # Freeze all backbone layers initially
+    # (we'll unfreeze later for fine-tuning)
+    for layer in base_model.layers:
+        if layer.name != "dense":  # We'll replace this anyway
+            layer.trainable = False
+
+    # Build new classification head
+    x = backbone_output
+    x = layers.Dropout(0.3, name="dropout_new")(x)
+    x = layers.Dense(
+        NUM_CLASSES,
+        activation="softmax",
+        name="dense_new",
+        kernel_regularizer=regularizers.l2(1e-4),
+    )(x)
+
+    # Create new model
+    model = keras.Model(
+        inputs=base_model.input, outputs=x, name="plant-disease-classifier-v2"
+    )
+
+    print(f"  New model input:  {model.input_shape}")
+    print(f"  New model output: {model.output_shape} ({NUM_CLASSES} classes)")
+
+    # Count trainable params
+    backbone_trainable = sum(
+        w.shape.num_elements()
+        for layer in base_model.layers
+        if layer.name != "dense"
+        for w in layer.trainable_weights
+    )
+    head_trainable = sum(
+        w.shape.num_elements() for w in model.get_layer("dense_new").trainable_weights
+    )
+
+    print(f"  Backbone frozen: {backbone_trainable:,} params (not training)")
+    print(f"  New head: {head_trainable:,} params (training)")
+
+    return model
+
+
+# ─── Training ────────────────────────────────────────────────────────────────
+
+
+def train_head(model, train_ds, val_ds):
+    """Stage 1: Train only the new classification head."""
+    print(f"\n{'=' * 60}")
+    print("STAGE 1: Training classification head")
+    print(f"{'=' * 60}")
+    print(f"  Epochs: {EPOCHS_HEAD}")
+    print(f"  Learning rate: {LEARNING_RATE_HEAD}")
+    print(f"  Batch size: {BATCH_SIZE}")
+
+    # Freeze all backbone layers
+    for layer in model.layers:
+        if layer.name != "dense_new":
+            layer.trainable = False
+        else:
+            layer.trainable = True
+
+    # Verify
+    trainable = sum(w.shape.num_elements() for w in model.trainable_weights)
+    total = sum(w.shape.num_elements() for w in model.weights)
+    print(f"  Trainable params: {trainable:,} / {total:,} total")
+
+    model.compile(
+        optimizer=optimizers.Adam(learning_rate=LEARNING_RATE_HEAD),
+        loss="sparse_categorical_crossentropy",
+        metrics=["accuracy", "sparse_top_k_categorical_accuracy"],
+    )
+
+    history = model.fit(
+        train_ds,
+        validation_data=val_ds,
+        epochs=EPOCHS_HEAD,
+        verbose=1,
+        callbacks=[
+            keras.callbacks.EarlyStopping(
+                monitor="val_accuracy",
+                patience=3,
+                restore_best_weights=True,
+            ),
+            keras.callbacks.ReduceLROnPlateau(
+                monitor="val_loss",
+                factor=0.5,
+                patience=2,
+                min_lr=1e-6,
+            ),
+        ],
+    )
+
+    final_val_acc = history.history["val_accuracy"][-1]
+    print(f"\n  Stage 1 complete! Val accuracy: {final_val_acc:.4f}")
+    return history
+
+
+def train_finetune(model, train_ds, val_ds):
+    """Stage 2: Unfreeze last ~25 layers and fine-tune."""
+    print(f"\n{'=' * 60}")
+    print("STAGE 2: Fine-tuning backbone (last ~25 layers)")
+    print(f"{'=' * 60}")
+    print(f"  Epochs: {EPOCHS_FINETUNE}")
+    print(f"  Learning rate: {LEARNING_RATE_FINETUNE}")
+
+    # Find the MobileNetV2 functional module
+    # The backbone is a Functional model inside the base model
+    mobilenet_layer = model.get_layer("mobilenetv2_1.00_160")
+
+    # Unfreeze the last ~25 layers of the backbone
+    total_backbone_layers = len(mobilenet_layer.layers)
+    unfreeze_from = max(0, total_backbone_layers - 25)
+    print(
+        f"  Backbone has {total_backbone_layers} layers, unfreezing from layer {unfreeze_from}"
+    )
+
+    for i, layer in enumerate(mobilenet_layer.layers):
+        layer.trainable = i >= unfreeze_from
+
+    # Also unfreeze the new head
+    model.get_layer("dense_new").trainable = True
+    model.get_layer("dropout_new").trainable = True
+
+    trainable = sum(w.shape.num_elements() for w in model.trainable_weights)
+    total = sum(w.shape.num_elements() for w in model.weights)
+    print(f"  Trainable params: {trainable:,} / {total:,} total")
+
+    model.compile(
+        optimizer=optimizers.Adam(learning_rate=LEARNING_RATE_FINETUNE),
+        loss="sparse_categorical_crossentropy",
+        metrics=["accuracy", "sparse_top_k_categorical_accuracy"],
+    )
+
+    history = model.fit(
+        train_ds,
+        validation_data=val_ds,
+        epochs=EPOCHS_FINETUNE,
+        verbose=1,
+        callbacks=[
+            keras.callbacks.EarlyStopping(
+                monitor="val_accuracy",
+                patience=3,
+                restore_best_weights=True,
+            ),
+            keras.callbacks.ReduceLROnPlateau(
+                monitor="val_loss",
+                factor=0.5,
+                patience=2,
+                min_lr=1e-7,
+            ),
+        ],
+    )
+
+    final_val_acc = history.history["val_accuracy"][-1]
+    print(f"\n  Stage 2 complete! Val accuracy: {final_val_acc:.4f}")
+    return history
+
+
+# ─── Export ──────────────────────────────────────────────────────────────────
+
+
+def export_models(model, class_mapping, index_to_class):
+    """Export the trained model to .keras and TF.js formats."""
+    print(f"\n{'=' * 60}")
+    print("EXPORTING")
+    print(f"{'=' * 60}")
+
+    # 1. Save as .keras (for future retraining)
+    keras_path = OUTPUT_DIR / "model-finetuned.keras"
+    model.save(str(keras_path))
+    print(f"  ✓ Saved .keras: {keras_path}")
+
+    # 2. Save class mapping alongside the model
+    mapping_path = OUTPUT_DIR / "class_mapping.json"
+    with open(mapping_path, "w") as f:
+        json.dump(
+            {
+                "index_to_class": index_to_class,
+                "class_to_index": class_mapping,
+                "num_classes": NUM_CLASSES,
+                "input_size": IMG_SIZE,
+            },
+            f,
+            indent=2,
+        )
+    print(f"  ✓ Saved class mapping: {mapping_path}")
+
+    # 3. Export to TF.js format
+    tfjs_path = str(TFJS_OUTPUT)
+    if TFJS_OUTPUT.exists():
+        shutil.rmtree(tfjs_path)
+
+    try:
+        import tensorflowjs as tfjs
+
+        tfjs.converters.save_keras_model(model, tfjs_path)
+        print(f"  ✓ Saved TF.js: {tfjs_path}/")
+        for f in sorted(TFJS_OUTPUT.iterdir()):
+            size = f.stat().st_size
+            print(f"      {f.name:<30s} {size:>10,} bytes")
+    except Exception as e:
+        print(f"  ⚠ TF.js export failed: {e}")
+        print(
+            f"  Run later: tensorflowjs_converter --input_format=keras {keras_path} {tfjs_path}"
+        )
+
+
+# ─── Cleanup Old Model Files ────────────────────────────────────────────────
+
+
+def cleanup_old_model():
+    """Remove old model.json and shards from the directory."""
+    for f in OUTPUT_DIR.glob("model.json"):
+        print(f"  Removing old: {f.name}")
+        f.unlink()
+    for f in OUTPUT_DIR.glob("group1-shard*"):
+        print(f"  Removing old: {f.name}")
+        f.unlink()
+
+
+# ─── Main ────────────────────────────────────────────────────────────────────
+
+
+def main():
+    print("=" * 60)
+    print("PLANT DISEASE MODEL FINE-TUNER")
+    print("=" * 60)
+
+    # 1. Build class mapping
+    print("\n[1/5] Building class mapping...")
+    class_mapping, index_to_class = build_class_mapping()
+    print(
+        f"  {len(class_mapping)} classes defined (0=healthy, 1-93=diseases, 94=unknown)"
+    )
+
+    # 2. Verify dataset
+    print("\n[2/5] Verifying dataset...")
+    if not DATASET_DIR.exists():
+        print(f"  ERROR: Dataset not found at {DATASET_DIR}")
+        print("  Run the scraper first: npx tsx scripts/scrape-training-dataset.ts")
+        sys.exit(1)
+
+    available, total = verify_dataset(class_mapping)
+    print_dataset_summary(available, total)
+
+    if total < 100:
+        print(f"  WARNING: Only {total} images. Consider scraping more data.")
+        print("  Continue anyway? (y/n)")
+        # Continue regardless — user can decide
+
+    # 3. Load dataset
+    print("\n[3/5] Loading and augmenting dataset...")
+    train_ds, val_ds = load_dataset(class_mapping, available)
+
+    # 4. Build and train model
+    print("\n[4/5] Building model...")
+    model = build_model()
+    model.summary()
+
+    # Check if training should run
+    if total > 0:
+        train_head(model, train_ds, val_ds)
+        train_finetune(model, train_ds, val_ds)
+
+        # 5. Export
+        print("\n[5/5] Exporting models...")
+        cleanup_old_model()
+        export_models(model, class_mapping, index_to_class)
+    else:
+        print("\n  Skipping training — no dataset available.")
+        sys.exit(1)
+
+    # ── Final Summary ────────────────────────────────────────────────────────
+
+    print(f"\n{'=' * 60}")
+    print("DONE! Model fine-tuned and exported.")
+    print(f"{'=' * 60}")
+    print("\nFiles created:")
+    print(f"  {OUTPUT_DIR / 'model-finetuned.keras'}")
+    print(f"  {OUTPUT_DIR / 'class_mapping.json'}")
+    print(f"  {TFJS_OUTPUT / 'model.json'}")
+    print("\nTo update your app:")
+    print("  1. Replace model files:")
+    print(f"     cp {TFJS_OUTPUT / 'model.json'} {OUTPUT_DIR / 'model.json'}")
+    print(f"     cp {TFJS_OUTPUT / 'group1-shard*'} {OUTPUT_DIR / '/'}")
+    print("  2. Restart the dev server")
+    print("  3. Test with: POST /api/identify")
+    print("\nNote: Update labels.ts if the class order changed.")
+
+
+if __name__ == "__main__":
+    main()

apps/web/scripts/scrape-training-dataset.ts

@@ -0,0 +1,660 @@
+#!/usr/bin/env node
+/**
+ * scrape-training-dataset.ts
+ *
+ * Collects a training dataset for fine-tuning by scraping DuckDuckGo image search.
+ *
+ * Targets:
+ *   - 200 images per disease class (93 diseases)
+ *   - 400 images for the "healthy" class
+ *   - Full resolution images stored in data/dataset/{class_id}/
+ *
+ * DuckDuckGo approach (no API key needed):
+ *   1. Fetch the main search page to extract a vqd (query) token
+ *   2. Use the vqd token to paginate through image results
+ *   3. Download each image to the dataset directory
+ *
+ * Usage: cd apps/web && npx tsx scripts/scrape-training-dataset.ts
+ *
+ * Progress is tracked in data/dataset/.progress.json — interrupt and resume safely.
+ */
+
+import "dotenv/config";
+import { readFileSync, writeFileSync, existsSync, mkdirSync, readdirSync, statSync } from "fs";
+import { resolve, extname, join } from "path";
+
+// ─── Config ─────────────────────────────────────────────────────────────────
+
+const DISEASES_JSON = resolve(__dirname, "../src/data/diseases.json");
+const PLANTS_JSON = resolve(__dirname, "../src/data/plants.json");
+
+const DATASET_DIR = resolve(__dirname, "../data/dataset");
+const PROGRESS_FILE = resolve(DATASET_DIR, ".progress.json");
+
+/** Target images per disease class */
+const TARGET_PER_DISEASE = 200;
+
+/** Target images for the "healthy" class (2× normal) */
+const TARGET_HEALTHY = 400;
+
+/** Delay between DuckDuckGo search API calls (ms) */
+const SEARCH_DELAY = 1500;
+
+/** Delay between image downloads (ms) */
+const DOWNLOAD_DELAY = 300;
+
+/** Max concurrent downloads */
+const CONCURRENT_DOWNLOADS = 5;
+
+/** Minimum image size in bytes to accept (reject tiny placeholders) */
+const MIN_IMAGE_SIZE = 10_000; // 10KB
+
+/** Maximum image size in bytes */
+const MAX_IMAGE_SIZE = 10 * 1024 * 1024; // 10MB
+
+/** Allowed image content types */
+const ALLOWED_CONTENT_TYPES = ["image/jpeg", "image/jpg", "image/png", "image/webp", "image/gif"];
+
+/** Allowed file extensions */
+const ALLOWED_EXTENSIONS = [".jpg", ".jpeg", ".png", ".webp"];
+
+/** User agent for requests */
+const UA =
+  "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36";
+
+// ─── Types ──────────────────────────────────────────────────────────────────
+
+interface DiseaseSeed {
+  id: string;
+  plantId: string;
+  name: string;
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  [key: string]: any;
+}
+
+interface PlantSeed {
+  id: string;
+  commonName: string;
+  scientificName: string;
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  [key: string]: any;
+}
+
+interface DuckDuckGoImageResult {
+  image: string;
+  title: string;
+  url: string;
+  thumbnail: string;
+  height: number;
+  width: number;
+}
+
+interface ClassProgress {
+  count: number;
+  downloaded: number;
+  failed: number;
+  skipped: number;
+  /** URLs we've already seen (to avoid duplicates) */
+  seenUrls: string[];
+  /** Whether we've exhausted search results */
+  exhausted: boolean;
+}
+
+interface Progress {
+  lastUpdated: string;
+  classes: Record<string, ClassProgress>;
+}
+
+/** Class ID for healthy plants */
+const HEALTHY_CLASS = "healthy";
+
+// ─── DuckDuckGo API ─────────────────────────────────────────────────────────
+
+/**
+ * Extract the vqd token from DuckDuckGo's search page.
+ * Required for paginating image results.
+ */
+async function getVqdToken(query: string): Promise<string> {
+  const url = `https://duckduckgo.com/?q=${encodeURIComponent(query)}&t=h_&iax=images&ia=images`;
+
+  const res = await fetch(url, {
+    headers: { "User-Agent": UA, Accept: "text/html" },
+    signal: AbortSignal.timeout(15_000),
+  });
+
+  if (!res.ok) {
+    throw new Error(`Failed to get vqd token: ${res.status}`);
+  }
+
+  const html = await res.text();
+
+  // Extract vqd token from the HTML
+  // Format: vqd='<token>' or vqd="<token>"
+  const match = html.match(/vqd['"]?\s*[:=]\s*['"]([a-f0-9-]+)['"]/);
+  if (!match) {
+    throw new Error(`Could not extract vqd token from DuckDuckGo response for "${query}"`);
+  }
+
+  return match[1];
+}
+
+/**
+ * Fetch a page of DuckDuckGo image results.
+ */
+async function searchImagesDuckDuckGo(
+  query: string,
+  vqd: string,
+  page: number,
+): Promise<DuckDuckGoImageResult[]> {
+  const url = `https://duckduckgo.com/i.js?q=${encodeURIComponent(query)}&vqd=${vqd}&o=json&p=${page}&f=,,,`;
+
+  const res = await fetch(url, {
+    headers: {
+      "User-Agent": UA,
+      Accept: "application/json",
+      Referer: `https://duckduckgo.com/?q=${encodeURIComponent(query)}&t=h_&iax=images&ia=images`,
+    },
+    signal: AbortSignal.timeout(15_000),
+  });
+
+  if (!res.ok) {
+    if (res.status === 429) {
+      console.warn("  ⚠ Rate limited (429). Waiting 10s...");
+      await sleep(10_000);
+      return searchImagesDuckDuckGo(query, vqd, page); // Retry
+    }
+    if (res.status === 403) {
+      console.warn("  ⚠ Forbidden (403). Token may have expired.");
+      return []; // Token expired — no more pages
+    }
+    throw new Error(`DuckDuckGo search failed: ${res.status}`);
+  }
+
+  const data = (await res.json()) as { results: DuckDuckGoImageResult[] };
+  return data.results ?? [];
+}
+
+/**
+ * Search DuckDuckGo images, automatically paginating to collect up to `target` results.
+ * Returns unique image URLs.
+ */
+async function collectImages(
+  query: string,
+  target: number,
+  seenUrls: Set<string>,
+): Promise<{ urls: string[]; exhausted: boolean }> {
+  const results: string[] = [];
+  let page = 1;
+  let exhausted = false;
+  let consecutiveEmpty = 0;
+
+  // Get vqd token
+  let vqd: string;
+  try {
+    vqd = await getVqdToken(query);
+  } catch (err) {
+    console.warn(`  ⚠ Failed to get vqd token: ${err instanceof Error ? err.message : "unknown"}`);
+    return { urls: [], exhausted: true };
+  }
+
+  while (results.length < target) {
+    await sleep(SEARCH_DELAY);
+
+    let pageResults: DuckDuckGoImageResult[];
+    try {
+      pageResults = await searchImagesDuckDuckGo(query, vqd, page);
+    } catch (err) {
+      console.warn(`  ⚠ Search error: ${err instanceof Error ? err.message : "unknown"}`);
+      break;
+    }
+
+    if (pageResults.length === 0) {
+      consecutiveEmpty++;
+      if (consecutiveEmpty >= 3) {
+        exhausted = true;
+        break;
+      }
+      page++;
+      continue;
+    }
+
+    consecutiveEmpty = 0;
+    let newCount = 0;
+
+    for (const r of pageResults) {
+      if (results.length >= target) break;
+
+      const imgUrl = r.image || r.url;
+
+      // Skip if we've already seen this URL
+      if (seenUrls.has(imgUrl)) continue;
+
+      // Validate URL looks like an image
+      const ext = extname(new URL(imgUrl).pathname).toLowerCase();
+      if (!ALLOWED_EXTENSIONS.includes(ext) && !ext) {
+        // No extension - still try, could be a CDN URL
+      }
+
+      seenUrls.add(imgUrl);
+      results.push(imgUrl);
+      newCount++;
+    }
+
+    if (newCount === 0 && pageResults.every((r) => seenUrls.has(r.image || r.url))) {
+      // All results on this page were already seen
+      page++;
+      continue;
+    }
+
+    if (results.length < target) {
+      page++;
+    }
+  }
+
+  return { urls: results.slice(0, target), exhausted };
+}
+
+// ─── Image Download ─────────────────────────────────────────────────────────
+
+/**
+ * Download a single image from a URL to the target path.
+ * Returns true if successful, false otherwise.
+ */
+async function downloadImage(url: string, destPath: string): Promise<boolean> {
+  try {
+    const res = await fetch(url, {
+      headers: { "User-Agent": UA, Accept: "image/webp,image/png,image/jpeg" },
+      signal: AbortSignal.timeout(15_000),
+    });
+
+    if (!res.ok) return false;
+
+    const contentType = res.headers.get("content-type") || "";
+    const contentLength = parseInt(res.headers.get("content-length") || "0", 10);
+
+    // Validate content type
+    if (!ALLOWED_CONTENT_TYPES.some((t) => contentType.includes(t))) {
+      return false;
+    }
+
+    // Validate size
+    if (contentLength > 0 && contentLength < MIN_IMAGE_SIZE) return false;
+    if (contentLength > MAX_IMAGE_SIZE) return false;
+
+    const buffer = Buffer.from(await res.arrayBuffer());
+
+    // Double-check actual buffer size
+    if (buffer.length < MIN_IMAGE_SIZE) return false;
+    if (buffer.length > MAX_IMAGE_SIZE) return false;
+
+    // Determine correct extension from content type or URL
+    let ext = extname(new URL(url).pathname).toLowerCase();
+    if (!ALLOWED_EXTENSIONS.includes(ext)) {
+      // Map from content type
+      if (contentType.includes("jpeg") || contentType.includes("jpg")) ext = ".jpg";
+      else if (contentType.includes("png")) ext = ".png";
+      else if (contentType.includes("webp")) ext = ".webp";
+      else ext = ".jpg"; // Default
+    }
+
+    const filePath = destPath.replace(/\.\w+$/, ext);
+    writeFileSync(filePath, buffer);
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+/**
+ * Download multiple images concurrently, respecting a per-download delay.
+ */
+async function downloadBatch(
+  urls: string[],
+  classDir: string,
+  startIndex: number,
+): Promise<{ downloaded: number; failed: number; lastIndex: number }> {
+  let downloaded = 0;
+  let failed = 0;
+  let index = startIndex;
+
+  // Process in chunks to control concurrency
+  for (let i = 0; i < urls.length; i += CONCURRENT_DOWNLOADS) {
+    const chunk = urls.slice(i, i + CONCURRENT_DOWNLOADS);
+
+    const results = await Promise.all(
+      chunk.map(async (url) => {
+        const paddedIndex = String(index).padStart(4, "0");
+        const destPath = resolve(classDir, `img_${paddedIndex}.jpg`);
+
+        const success = await downloadImage(url, destPath);
+        await sleep(DOWNLOAD_DELAY);
+        return { success, index: index++ };
+      }),
+    );
+
+    for (const r of results) {
+      if (r.success) downloaded++;
+      else failed++;
+    }
+  }
+
+  return { downloaded, failed, lastIndex: index };
+}
+
+// ─── Progress Tracking ──────────────────────────────────────────────────────
+
+function loadProgress(): Progress {
+  if (!existsSync(PROGRESS_FILE)) {
+    return { lastUpdated: new Date().toISOString(), classes: {} };
+  }
+  return JSON.parse(readFileSync(PROGRESS_FILE, "utf-8")) as Progress;
+}
+
+function saveProgress(progress: Progress): void {
+  progress.lastUpdated = new Date().toISOString();
+  writeFileSync(PROGRESS_FILE, JSON.stringify(progress, null, 2));
+}
+
+function getClassProgress(progress: Progress, classId: string): ClassProgress {
+  if (!progress.classes[classId]) {
+    progress.classes[classId] = {
+      count: 0,
+      downloaded: 0,
+      failed: 0,
+      skipped: 0,
+      seenUrls: [],
+      exhausted: false,
+    };
+  }
+  return progress.classes[classId];
+}
+
+// ─── Search Query Building ──────────────────────────────────────────────────
+
+function buildSearchQueries(disease: DiseaseSeed, plant: PlantSeed | null): string[] {
+  const name = disease.name;
+  const plantName = plant?.commonName || disease.plantId;
+
+  return [
+    `${name} ${plantName} leaf disease`,
+    `${plantName} ${name} symptoms`,
+    `${name} plant disease`,
+    `${plantName} diseased leaf`,
+  ];
+}
+
+function buildHealthyQueries(plant: PlantSeed): string[] {
+  return [
+    `healthy ${plant.commonName} leaf`,
+    `${plant.commonName} leaf closeup`,
+    `healthy ${plant.commonName} plant`,
+    `${plant.commonName} foliage`,
+  ];
+}
+
+// ─── Dataset Collection ─────────────────────────────────────────────────────
+
+async function collectClassImages(
+  classId: string,
+  queries: string[],
+  target: number,
+  progress: Progress,
+  classDir: string,
+): Promise<void> {
+  const cp = getClassProgress(progress, classId);
+  const seenUrls = new Set(cp.seenUrls);
+
+  if (cp.count >= target) {
+    console.log(`  ✓ Already have ${cp.count}/${target} images`);
+    return;
+  }
+
+  if (cp.exhausted) {
+    console.log(`  ✓ Already exhausted search results (${cp.count}/${target} images)`);
+    return;
+  }
+
+  mkdirSync(classDir, { recursive: true });
+
+  const totalUrls: string[] = [];
+  let exhausted = false;
+
+  // Search with each query until we hit the target
+  for (const query of queries) {
+    if (totalUrls.length >= target) break;
+
+    console.log(`  Searching: "${query}"...`);
+    const result = await collectImages(query, target - totalUrls.length, seenUrls);
+
+    totalUrls.push(...result.urls);
+    cp.seenUrls = Array.from(seenUrls);
+
+    if (result.exhausted) {
+      exhausted = true;
+    }
+
+    if (totalUrls.length >= target) break;
+  }
+
+  if (totalUrls.length === 0) {
+    cp.exhausted = exhausted;
+    saveProgress(progress);
+    console.log(`  ✗ No images found for "${classId}"`);
+    return;
+  }
+
+  console.log(`  Found ${totalUrls.length} unique image URLs. Downloading...`);
+
+  // Download the images
+  const { downloaded, failed } = await downloadBatch(totalUrls, classDir, cp.count);
+
+  cp.count += downloaded;
+  cp.downloaded += downloaded;
+  cp.failed += failed;
+  cp.exhausted = exhausted;
+
+  saveProgress(progress);
+
+  const pct = Math.round((cp.count / target) * 100);
+  console.log(
+    `  ${downloaded > 0 ? "✓" : "✗"} Got ${downloaded} images (${failed} failed). Total: ${cp.count}/${target} (${pct}%)`,
+  );
+}
+
+// ─── Main ───────────────────────────────────────────────────────────────────
+
+async function main() {
+  console.log("=".repeat(60));
+  console.log("PLANT DISEASE DATASET COLLECTOR");
+  console.log("=".repeat(60));
+
+  // Load knowledge base
+  const diseases = JSON.parse(readFileSync(DISEASES_JSON, "utf-8")) as DiseaseSeed[];
+  const plants = JSON.parse(readFileSync(PLANTS_JSON, "utf-8")) as PlantSeed[];
+
+  const plantMap = new Map<string, PlantSeed>();
+  for (const p of plants) {
+    plantMap.set(p.id, p);
+  }
+
+  console.log(`\nLoaded ${diseases.length} diseases, ${plants.length} plants`);
+  console.log(
+    `Target: ${TARGET_PER_DISEASE} images/disease (×${diseases.length} = ${diseases.length * TARGET_PER_DISEASE})`,
+  );
+  console.log(`Target: ${TARGET_HEALTHY} images for "healthy" class`);
+  console.log(`Output: ${DATASET_DIR}/`);
+  console.log("");
+
+  // Load progress
+  mkdirSync(DATASET_DIR, { recursive: true });
+  const progress = loadProgress();
+
+  const startTime = Date.now();
+
+  // ── Phase 1: Disease classes ──────────────────────────────────────────────
+
+  console.log("─".repeat(60));
+  console.log("PHASE 1: Disease Images");
+  console.log("─".repeat(60));
+
+  for (let i = 0; i < diseases.length; i++) {
+    const disease = diseases[i];
+    const plant = plantMap.get(disease.plantId) ?? null;
+    const classDir = resolve(DATASET_DIR, disease.id);
+    const queries = buildSearchQueries(disease, plant);
+
+    const pct = Math.round((i / diseases.length) * 100);
+    console.log(`\n[${i + 1}/${diseases.length}] (${pct}%) ${disease.name} (${disease.id})`);
+
+    await collectClassImages(disease.id, queries, TARGET_PER_DISEASE, progress, classDir);
+  }
+
+  // ── Phase 2: Healthy class ────────────────────────────────────────────────
+
+  console.log("\n" + "─".repeat(60));
+  console.log("PHASE 2: Healthy Plant Images");
+  console.log("─".repeat(60));
+
+  const healthyDir = resolve(DATASET_DIR, HEALTHY_CLASS);
+  const healthyCp = getClassProgress(progress, HEALTHY_CLASS);
+  const healthySeen = new Set(healthyCp.seenUrls);
+
+  if (healthyCp.count >= TARGET_HEALTHY) {
+    console.log(`\n  ✓ Already have ${healthyCp.count}/${TARGET_HEALTHY} healthy images`);
+  } else {
+    // Build a pool of healthy plant queries
+    const allHealthyQueries: string[] = [];
+    for (const plant of plants) {
+      allHealthyQueries.push(...buildHealthyQueries(plant));
+    }
+
+    const totalHealthyUrls: string[] = [];
+    let healthyExhausted = false;
+
+    for (const query of allHealthyQueries) {
+      if (totalHealthyUrls.length >= TARGET_HEALTHY) break;
+      if (healthyExhausted) break;
+
+      console.log(`\n  Searching: "${query}"...`);
+      const result = await collectImages(
+        query,
+        TARGET_HEALTHY - totalHealthyUrls.length,
+        healthySeen,
+      );
+
+      totalHealthyUrls.push(...result.urls);
+
+      if (result.exhausted) {
+        healthyExhausted = true;
+      }
+    }
+
+    healthyCp.seenUrls = Array.from(healthySeen);
+
+    if (totalHealthyUrls.length > 0) {
+      console.log(`\n  Found ${totalHealthyUrls.length} healthy image URLs. Downloading...`);
+      const { downloaded, failed } = await downloadBatch(
+        totalHealthyUrls,
+        healthyDir,
+        healthyCp.count,
+      );
+
+      healthyCp.count += downloaded;
+      healthyCp.downloaded += downloaded;
+      healthyCp.failed += failed;
+      healthyCp.exhausted = healthyExhausted;
+
+      const pct = Math.round((healthyCp.count / TARGET_HEALTHY) * 100);
+      console.log(
+        `  Got ${downloaded} images (${failed} failed). Total: ${healthyCp.count}/${TARGET_HEALTHY} (${pct}%)`,
+      );
+    } else {
+      healthyCp.exhausted = true;
+      console.log(`  ✗ No healthy images found`);
+    }
+
+    saveProgress(progress);
+  }
+
+  // ── Summary ────────────────────────────────────────────────────────────────
+
+  const elapsed = Math.round((Date.now() - startTime) / 1000);
+  const mins = Math.floor(elapsed / 60);
+  const secs = elapsed % 60;
+
+  let totalDownloaded = 0;
+  let totalFailed = 0;
+  let totalTarget = 0;
+
+  for (const [classId, cp] of Object.entries(progress.classes)) {
+    totalDownloaded += cp.downloaded || 0;
+    totalFailed += cp.failed || 0;
+    totalTarget += classId === HEALTHY_CLASS ? TARGET_HEALTHY : TARGET_PER_DISEASE;
+  }
+
+  const totalSize = await getDatasetSize();
+  const sizeGb = (totalSize / (1024 * 1024 * 1024)).toFixed(2);
+
+  console.log("\n" + "=".repeat(60));
+  console.log("COMPLETE");
+  console.log("=".repeat(60));
+  console.log(`  Time:       ${mins}m ${secs}s`);
+  console.log(`  Downloaded: ${totalDownloaded} images`);
+  console.log(`  Failed:     ${totalFailed} images`);
+  console.log(`  Target:     ${totalTarget} images`);
+  console.log(`  Dataset size: ${sizeGb} GB`);
+  console.log(`  Dataset location: ${DATASET_DIR}/`);
+  console.log("");
+  console.log("Next steps:");
+  console.log("  1. Run the fine-tuning script to train on this dataset");
+  console.log("  2. The fine-tuning script will resize to 160×160 and augment");
+  console.log("=".repeat(60));
+}
+
+/**
+ * Calculate total size of the dataset directory.
+ */
+async function getDatasetSize(): Promise<number> {
+  let total = 0;
+  if (!existsSync(DATASET_DIR)) return 0;
+
+  const entries = readdirSync(DATASET_DIR, { withFileTypes: true });
+
+  for (const entry of entries) {
+    if (!entry.name.startsWith(".")) {
+      const fullPath = resolve(DATASET_DIR, entry.name);
+      if (entry.isDirectory()) {
+        total += dirSize(fullPath);
+      }
+    }
+  }
+
+  return total;
+}
+
+function dirSize(dirPath: string): number {
+  let total = 0;
+  try {
+    const entries = readdirSync(dirPath, { withFileTypes: true });
+    for (const entry of entries) {
+      const fullPath = join(dirPath, entry.name);
+      if (entry.isFile()) {
+        total += statSync(fullPath).size;
+      } else if (entry.isDirectory()) {
+        total += dirSize(fullPath);
+      }
+    }
+  } catch {
+    // skip errors
+  }
+  return total;
+}
+
+function sleep(ms: number): Promise<void> {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+
+main().catch((err) => {
+  console.error("Fatal error:", err);
+  process.exit(1);
+});

apps/web/src/lib/image-processing.test.ts

@@ -173,14 +173,14 @@ describe("imageToTensor", () => {
 
 describe("tensorToBase64 / base64ToTensor", () => {
   it("round-trips tensor data correctly", () => {
-    const imageData = createMockImageData(224, 224, 100, 150, 200);
+    const imageData = createMockImageData(160, 160, 100, 150, 200);
     const original = imageToTensor(imageData);
 
     const base64 = tensorToBase64(original);
     const decoded = base64ToTensor(base64);
 
     expect(decoded.tensor.length).toBe(original.length);
-    expect(decoded.shape).toEqual([3, 224, 224]);
+    expect(decoded.shape).toEqual([3, 160, 160]);
 
     // Check a few values match
     for (let i = 0; i < 10; i++) {
@@ -197,9 +197,9 @@ describe("tensorToBase64 / base64ToTensor", () => {
 });
 
 describe("getTensorShape", () => {
-  it("returns [1, 3, 224, 224] by default", () => {
+  it("returns [1, 3, 160, 160] by default", () => {
     const shape = getTensorShape();
-    expect(shape).toEqual([1, 3, 224, 224]);
+    expect(shape).toEqual([1, 3, 160, 160]);
   });
 
   it("returns NCHW layout", () => {
@@ -207,8 +207,8 @@ describe("getTensorShape", () => {
     expect(shape.length).toBe(4);
     expect(shape[0]).toBe(1); // batch
     expect(shape[1]).toBe(3); // channels
-    expect(shape[2]).toBe(224); // height
+    expect(shape[2]).toBe(160); // height (model input size)
-    expect(shape[3]).toBe(224); // width
+    expect(shape[3]).toBe(160); // width (model input size)
   });
 });
 

apps/web/src/lib/image-processing.ts

@@ -17,7 +17,7 @@
 
 const DEFAULT_MODEL_SIZE = 160;
 const DEFAULT_MEAN = [0.485, 0.456, 0.406] as const; // ImageNet RGB means
-const DEFAULT_STD = [0.229, 0.224, 0.225] as const;  // ImageNet RGB stds
+const DEFAULT_STD = [0.229, 0.224, 0.225] as const; // ImageNet RGB stds
 
 function getConfig(): {
   size: number;
@@ -48,12 +48,7 @@ export const MAX_FILE_SIZE = 10 * 1024 * 1024;
 export const MIN_DIMENSION = 150;
 
 /** Allowed MIME types */
-export const ALLOWED_MIME_TYPES = [
+export const ALLOWED_MIME_TYPES = ["image/png", "image/jpeg", "image/jpg", "image/webp"] as const;
-  "image/png",
-  "image/jpeg",
-  "image/jpg",
-  "image/webp",
-] as const;
 
 export type AllowedMimeType = (typeof ALLOWED_MIME_TYPES)[number];
 
@@ -66,9 +61,7 @@ export const MAX_UPLOADS = 100;
  * Validate that a file is an acceptable image for upload.
  * Returns `{ ok: true }` or `{ ok: false, error: string }`.
  */
-export function validateImageFile(file: File):
+export function validateImageFile(file: File): { ok: true } | { ok: false; error: string } {
-  | { ok: true }
-  | { ok: false; error: string } {
   // MIME type check
   if (!ALLOWED_MIME_TYPES.includes(file.type as AllowedMimeType)) {
     return {
@@ -127,10 +120,7 @@ export function validateImageDimensions(
  * @param size - Target dimension (square). Defaults to IMAGE_MODEL_SIZE env or 224.
  * @returns ImageData at exactly `size × size`
  */
-export async function resizeImage(
+export async function resizeImage(file: File, size: number = getConfig().size): Promise<ImageData> {
-  file: File,
-  size: number = getConfig().size,
-): Promise<ImageData> {
   return new Promise((resolve, reject) => {
     const img = new Image();
     img.onload = () => {
@@ -193,8 +183,7 @@ export function imageToTensor(imageData: ImageData): Float32Array {
 
   // Normalize with ImageNet mean/std
   for (let c = 0; c < 3; c++) {
-    const channel =
+    const channel = c === 0 ? rChannel : c === 1 ? gChannel : bChannel;
-      c === 0 ? rChannel : c === 1 ? gChannel : bChannel;
     const m = mean[c];
     const s = std[c];
     for (let i = 0; i < totalPixels; i++) {
@@ -253,5 +242,3 @@ export function base64ToTensor(base64: string): {
     shape: envelope.shape as [number, number, number],
   };
 }
-
-

apps/web/src/lib/ml/inference.test.ts

@@ -97,7 +97,7 @@ describe("createZeroTensor", () => {
 
   it("all values are zero", () => {
     const tensor = createZeroTensor();
-    expect(tensor.every(v => v === 0)).toBe(true);
+    expect(tensor.every((v) => v === 0)).toBe(true);
   });
 });
 
@@ -114,12 +114,12 @@ describe("createRandomTensor", () => {
 
   it("all values are finite", () => {
     const tensor = createRandomTensor();
-    expect(tensor.every(v => Number.isFinite(v))).toBe(true);
+    expect(tensor.every((v) => Number.isFinite(v))).toBe(true);
   });
 
   it("produces varied values", () => {
     const tensor = createRandomTensor();
-    const uniqueValues = new Set(tensor.map(v => v.toFixed(4)));
+    const uniqueValues = new Set(tensor.map((v) => v.toFixed(4)));
     expect(uniqueValues.size).toBeGreaterThan(100);
   });
 
@@ -172,7 +172,7 @@ describe("runInference", () => {
     const result = await runInference(tensor);
     for (let i = 0; i < result.predictions.length - 1; i++) {
       expect(result.predictions[i].probability).toBeGreaterThanOrEqual(
-        result.predictions[i + 1].probability
+        result.predictions[i + 1].probability,
       );
     }
   }, 10000);

apps/web/src/lib/ml/inference.ts

@@ -69,9 +69,7 @@ export async function runInference(
  */
 export function validateInput(tensor: Float32Array): void {
   if (!(tensor instanceof Float32Array)) {
-    throw new Error(
+    throw new Error(`Expected Float32Array input, got ${typeof tensor}`);
-      `Expected Float32Array input, got ${typeof tensor}`,
-    );
   }
 
   if (tensor.length !== INPUT_SIZE) {
@@ -84,9 +82,7 @@ export function validateInput(tensor: Float32Array): void {
   // Check for NaN/Infinity values
   for (let i = 0; i < tensor.length; i++) {
     if (!Number.isFinite(tensor[i])) {
-      throw new Error(
+      throw new Error(`Tensor contains non-finite value at index ${i}: ${tensor[i]}`);
-        `Tensor contains non-finite value at index ${i}: ${tensor[i]}`,
-      );
     }
   }
 }

apps/web/src/lib/ml/labels.test.ts

@@ -1,17 +1,21 @@
 /**
  * Unit tests for lib/ml/labels.ts
  *
- * Tests:
+ * The model has 38 PlantVillage classes. Some map to the app's
- * - INDEX_TO_DISEASE_ID maps index 0 to "healthy"
+ * knowledge base disease IDs, others map to "unknown".
- * - INDEX_TO_DISEASE_ID maps last index to "unknown"
+ *
- * - INDEX_TO_DISEASE_ID maps intermediate indices to disease IDs
+ * Known mappings:
- * - DISEASE_ID_TO_INDEX is inverse of INDEX_TO_DISEASE_ID
+ * - indices 3, 4, 6, 10, 14, 17, 19, 22, 23, 24, 27, 37 → "healthy"
- * - getDiseaseIdForIndex returns "unknown" for out-of-range
+ * - index 20 (Potato___Early_blight) → "early-blight"
- * - getIndexForDiseaseId returns -1 for unknown ID
+ * - index 21 (Potato___Late_blight) → "late-blight"
- * - isRealDisease correctly classifies healthy/unknown vs real diseases
+ * - index 25 (Squash___Powdery_mildew) → "squash-powdery-mildew"
- * - getAllDiseaseIds returns all disease IDs from knowledge base
+ * - index 26 (Strawberry___Leaf_scorch) → "strawberry-leaf-scorch"
- * - NUM_CLASSES equals expected count (diseases + healthy + unknown)
+ * - index 28 (Tomato___Bacterial_spot) → "bacterial-leaf-spot-tomato"
- * - Bidirectional mapping consistency
+ * - index 29 (Tomato___Early_blight) → "early-blight" (duplicate)
+ * - index 30 (Tomato___Late_blight) → "late-blight" (duplicate)
+ * - index 32 (Tomato___Septoria_leaf_spot) → "septoria-leaf-spot"
+ * - index 37 (Tomato___healthy) → "healthy"
+ * - all others → "unknown"
  */
 
 import { describe, it, expect } from "vitest";
@@ -23,143 +27,105 @@ import {
   isRealDisease,
   getAllDiseaseIds,
   NUM_CLASSES,
-  HEALTHY_INDEX,
+  getPlantVillageClassName,
-  FIRST_DISEASE_INDEX,
-  UNKNOWN_INDEX,
 } from "@/lib/ml/labels";
-import rawDiseases from "@/data/diseases.json";
-import type { Disease } from "@/lib/types";
-
-const diseases: Disease[] = rawDiseases as Disease[];
 
 describe("Constants", () => {
-  it("HEALTHY_INDEX is 0", () => {
+  it("NUM_CLASSES is 38 (PlantVillage)", () => {
-    expect(HEALTHY_INDEX).toBe(0);
+    expect(NUM_CLASSES).toBe(38);
   });
 
-  it("FIRST_DISEASE_INDEX is 1", () => {
+  it("all 38 indices are mapped", () => {
-    expect(FIRST_DISEASE_INDEX).toBe(1);
+    const keys = Object.keys(INDEX_TO_DISEASE_ID).map(Number);
-  });
+    expect(keys.length).toBe(38);
-
+    for (let i = 0; i < 38; i++) {
-  it("UNKNOWN_INDEX is 1 + number of diseases", () => {
+      expect(keys).toContain(i);
-    expect(UNKNOWN_INDEX).toBe(1 + diseases.length);
+    }
-  });
-
-  it("NUM_CLASSES is UNKNOWN_INDEX + 1", () => {
-    expect(NUM_CLASSES).toBe(UNKNOWN_INDEX + 1);
-  });
-
-  it("NUM_CLASSES equals diseases.length + 2 (healthy + unknown)", () => {
-    expect(NUM_CLASSES).toBe(diseases.length + 2);
   });
 });
 
-describe("INDEX_TO_DISEASE_ID", () => {
+describe("INDEX_TO_DISEASE_ID — healthy indices", () => {
-  it("maps index 0 to 'healthy'", () => {
+  const healthyIndices = [3, 4, 6, 10, 14, 17, 19, 22, 23, 24, 27, 37];
-    expect(INDEX_TO_DISEASE_ID[0]).toBe("healthy");
-  });
 
-  it("maps last index to 'unknown'", () => {
+  for (const idx of healthyIndices) {
-    expect(INDEX_TO_DISEASE_ID[NUM_CLASSES - 1]).toBe("unknown");
+    it(`index ${idx} maps to "healthy"`, () => {
-  });
+      expect(INDEX_TO_DISEASE_ID[idx]).toBe("healthy");
+    });
+  }
+});
 
-  it("maps intermediate indices to disease IDs", () => {
+describe("INDEX_TO_DISEASE_ID — known disease mappings", () => {
-    // Index 1 should be the first disease
+  const cases: Array<{ index: number; expected: string; name: string }> = [
-    expect(INDEX_TO_DISEASE_ID[1]).toBe(diseases[0].id);
+    { index: 20, expected: "early-blight", name: "Potato___Early_blight" },
-    // Index 2 should be the second disease
+    { index: 21, expected: "late-blight", name: "Potato___Late_blight" },
-    expect(INDEX_TO_DISEASE_ID[2]).toBe(diseases[1].id);
+    { index: 25, expected: "squash-powdery-mildew", name: "Squash___Powdery_mildew" },
-    // Last disease index
+    { index: 26, expected: "strawberry-leaf-scorch", name: "Strawberry___Leaf_scorch" },
-    expect(INDEX_TO_DISEASE_ID[diseases.length]).toBe(diseases[diseases.length - 1].id);
+    { index: 28, expected: "bacterial-leaf-spot-tomato", name: "Tomato___Bacterial_spot" },
-  });
+    { index: 29, expected: "early-blight", name: "Tomato___Early_blight" },
+    { index: 30, expected: "late-blight", name: "Tomato___Late_blight" },
+    { index: 32, expected: "septoria-leaf-spot", name: "Tomato___Septoria_leaf_spot" },
+  ];
 
-  it("has exactly NUM_CLASSES entries", () => {
+  for (const { index, expected, name } of cases) {
-    const keys = Object.keys(INDEX_TO_DISEASE_ID);
+    it(`index ${index} (${name}) maps to "${expected}"`, () => {
-    expect(keys.length).toBe(NUM_CLASSES);
+      expect(INDEX_TO_DISEASE_ID[index]).toBe(expected);
-  });
+    });
+  }
+});
 
-  it("all mapped IDs are valid strings", () => {
+describe("INDEX_TO_DISEASE_ID — unknown (unmapped) indices", () => {
-    for (const id of Object.values(INDEX_TO_DISEASE_ID)) {
+  const unknownIndices = [0, 1, 2, 5, 7, 8, 9, 11, 12, 13, 15, 16, 18, 31, 33, 34, 35, 36];
-      expect(typeof id).toBe("string");
+
-      expect(id.length).toBeGreaterThan(0);
+  for (const idx of unknownIndices) {
-    }
+    it(`index ${idx} maps to "unknown"`, () => {
-  });
+      expect(INDEX_TO_DISEASE_ID[idx]).toBe("unknown");
+    });
+  }
 });
 
 describe("DISEASE_ID_TO_INDEX", () => {
-  it("maps 'healthy' to index 0", () => {
+  it("maps 'early-blight' to first occurrence (index 20)", () => {
-    expect(DISEASE_ID_TO_INDEX["healthy"]).toBe(0);
+    expect(DISEASE_ID_TO_INDEX["early-blight"]).toBe(20);
   });
 
-  it("maps 'unknown' to last index", () => {
+  it("maps 'late-blight' to first occurrence (index 21)", () => {
-    expect(DISEASE_ID_TO_INDEX["unknown"]).toBe(NUM_CLASSES - 1);
+    expect(DISEASE_ID_TO_INDEX["late-blight"]).toBe(21);
   });
 
-  it("maps disease IDs to correct indices", () => {
+  it("maps 'septoria-leaf-spot' to index 32", () => {
-    for (let i = 0; i < diseases.length; i++) {
+    expect(DISEASE_ID_TO_INDEX["septoria-leaf-spot"]).toBe(32);
-      expect(DISEASE_ID_TO_INDEX[diseases[i].id]).toBe(FIRST_DISEASE_INDEX + i);
-    }
   });
 
-  it("has exactly NUM_CLASSES entries", () => {
+  it("maps 'healthy' to index 3 (first healthy index)", () => {
-    const keys = Object.keys(DISEASE_ID_TO_INDEX);
+    expect(DISEASE_ID_TO_INDEX["healthy"]).toBe(3);
-    expect(keys.length).toBe(NUM_CLASSES);
   });
 });
 
 describe("Bidirectional mapping", () => {
-  it("INDEX_TO_DISEASE_ID and DISEASE_ID_TO_INDEX are inverses", () => {
+  it("every index round-trips correctly", () => {
-    for (const [idxStr, id] of Object.entries(INDEX_TO_DISEASE_ID)) {
-      const idx = parseInt(idxStr);
-      expect(DISEASE_ID_TO_INDEX[id]).toBe(idx);
-    }
-  });
-
-  it("round-trips for all disease IDs", () => {
-    for (const [id, idx] of Object.entries(DISEASE_ID_TO_INDEX)) {
-      expect(INDEX_TO_DISEASE_ID[idx]).toBe(id);
-    }
-  });
-
-  it("round-trips for all indices", () => {
     for (let i = 0; i < NUM_CLASSES; i++) {
       const id = INDEX_TO_DISEASE_ID[i];
-      expect(DISEASE_ID_TO_INDEX[id]).toBe(i);
+      const idx = DISEASE_ID_TO_INDEX[id];
+      expect(INDEX_TO_DISEASE_ID[idx]).toBe(id);
     }
   });
 });
 
 describe("getDiseaseIdForIndex", () => {
-  it("returns 'healthy' for index 0", () => {
-    expect(getDiseaseIdForIndex(0)).toBe("healthy");
-  });
-
-  it("returns disease ID for valid disease index", () => {
-    expect(getDiseaseIdForIndex(1)).toBe(diseases[0].id);
-  });
-
   it("returns 'unknown' for out-of-range positive index", () => {
-    expect(getDiseaseIdForIndex(1000)).toBe("unknown");
+    expect(getDiseaseIdForIndex(100)).toBe("unknown");
   });
 
   it("returns 'unknown' for negative index", () => {
     expect(getDiseaseIdForIndex(-1)).toBe("unknown");
   });
 
-  it("returns 'unknown' for index past NUM_CLASSES", () => {
+  it("returns correct ID for valid index", () => {
-    expect(getDiseaseIdForIndex(NUM_CLASSES + 10)).toBe("unknown");
+    expect(getDiseaseIdForIndex(20)).toBe("early-blight");
   });
 });
 
 describe("getIndexForDiseaseId", () => {
-  it("returns 0 for 'healthy'", () => {
-    expect(getIndexForDiseaseId("healthy")).toBe(0);
-  });
-
-  it("returns correct index for known disease", () => {
-    const idx = getIndexForDiseaseId(diseases[0].id);
-    expect(idx).toBe(1);
-  });
-
   it("returns -1 for unknown disease ID", () => {
     expect(getIndexForDiseaseId("nonexistent-disease")).toBe(-1);
   });
@@ -169,9 +135,7 @@ describe("getIndexForDiseaseId", () => {
   });
 
   it("is case-insensitive", () => {
-    const lowerIdx = getIndexForDiseaseId(diseases[0].id);
+    expect(getIndexForDiseaseId("EARLY-BLIGHT")).toBe(20);
-    const upperIdx = getIndexForDiseaseId(diseases[0].id.toUpperCase());
-    expect(upperIdx).toBe(lowerIdx);
   });
 });
 
@@ -184,10 +148,9 @@ describe("isRealDisease", () => {
     expect(isRealDisease("unknown")).toBe(false);
   });
 
-  it("returns true for actual disease IDs", () => {
+  it("returns true for known disease IDs", () => {
-    for (const disease of diseases) {
+    expect(isRealDisease("early-blight")).toBe(true);
-      expect(isRealDisease(disease.id)).toBe(true);
+    expect(isRealDisease("septoria-leaf-spot")).toBe(true);
-    }
   });
 
   it("returns true for arbitrary non-special strings", () => {
@@ -195,27 +158,37 @@ describe("isRealDisease", () => {
   });
 });
 
+describe("getPlantVillageClassName", () => {
+  it("returns correct class name for tomato healthy", () => {
+    expect(getPlantVillageClassName(37)).toBe("Tomato___healthy");
+  });
+
+  it("returns correct class name for potato early blight", () => {
+    expect(getPlantVillageClassName(20)).toBe("Potato___Early_blight");
+  });
+
+  it("returns 'unknown' for out-of-range index", () => {
+    expect(getPlantVillageClassName(100)).toBe("unknown");
+  });
+});
+
 describe("getAllDiseaseIds", () => {
-  it("returns array of all disease IDs", () => {
+  it("returns only mapped disease IDs", () => {
     const ids = getAllDiseaseIds();
-    expect(ids.length).toBe(diseases.length);
+    expect(ids).toContain("early-blight");
+    expect(ids).toContain("late-blight");
+    expect(ids).toContain("squash-powdery-mildew");
+    expect(ids).toContain("strawberry-leaf-scorch");
+    expect(ids).toContain("bacterial-leaf-spot-tomato");
+    expect(ids).toContain("septoria-leaf-spot");
   });
 
   it("excludes 'healthy'", () => {
-    const ids = getAllDiseaseIds();
+    expect(getAllDiseaseIds()).not.toContain("healthy");
-    expect(ids).not.toContain("healthy");
   });
 
   it("excludes 'unknown'", () => {
-    const ids = getAllDiseaseIds();
+    expect(getAllDiseaseIds()).not.toContain("unknown");
-    expect(ids).not.toContain("unknown");
-  });
-
-  it("includes all disease IDs from knowledge base", () => {
-    const ids = getAllDiseaseIds();
-    for (const disease of diseases) {
-      expect(ids).toContain(disease.id);
-    }
   });
 
   it("has no duplicates", () => {

apps/web/src/lib/ml/labels.ts

@@ -1,74 +1,197 @@
 /**
  * Class label mapping for the plant disease classifier model.
  *
- * Maps model output index → disease ID string.
+ * This model is a MobileNetV2 trained on the PlantVillage dataset
- * The model has classes for each disease in the knowledge base,
+ * with 38 classes (14 crops × diseases/healthy).
- * plus "healthy" and "unknown" catch-all classes.
  *
- * Model output shape: [1, NUM_CLASSES] where NUM_CLASSES = 95
+ * Model output shape: [1, NUM_CLASSES] where NUM_CLASSES = 38
- * (93 diseases + "healthy" + "unknown")
  *
- * Index layout:
+ * Index layout (from labels_pv_original.json):
- *   0  → "healthy"
+ *   0  → Apple___Apple_scab
- *   1–93 → disease IDs (order matches diseases.json)
+ *   1  → Apple___Black_rot
- *   94 → "unknown"
+ *   2  → Apple___Cedar_apple_rust
+ *   3  → Apple___healthy
+ *   4  → Blueberry___healthy
+ *   5  → Cherry_(including_sour)___Powdery_mildew
+ *   6  → Cherry_(including_sour)___healthy
+ *   7  → Corn_(maize)___Cercospora_leaf_spot Gray_leaf_spot
+ *   8  → Corn_(maize)___Common_rust_
+ *   9  → Corn_(maize)___Northern_Leaf_Blight
+ *   10 → Corn_(maize)___healthy
+ *   11 → Grape___Black_rot
+ *   12 → Grape___Esca_(Black_Measles)
+ *   13 → Grape___Leaf_blight_(Isariopsis_Leaf_Spot)
+ *   14 → Grape___healthy
+ *   15 → Orange___Haunglongbing_(Citrus_greening)
+ *   16 → Peach___Bacterial_spot
+ *   17 → Peach___healthy
+ *   18 → Pepper,_bell___Bacterial_spot
+ *   19 → Pepper,_bell___healthy
+ *   20 → Potato___Early_blight
+ *   21 → Potato___Late_blight
+ *   22 → Potato___healthy
+ *   23 → Raspberry___healthy
+ *   24 → Soybean___healthy
+ *   25 → Squash___Powdery_mildew
+ *   26 → Strawberry___Leaf_scorch
+ *   27 → Strawberry___healthy
+ *   28 → Tomato___Bacterial_spot
+ *   29 → Tomato___Early_blight
+ *   30 → Tomato___Late_blight
+ *   31 → Tomato___Leaf_Mold
+ *   32 → Tomato___Septoria_leaf_spot
+ *   33 → Tomato___Spider_mites Two-spotted_spider_mite
+ *   34 → Tomato___Target_Spot
+ *   35 → Tomato___Tomato_Yellow_Leaf_Curl_Virus
+ *   36 → Tomato___Tomato_mosaic_virus
+ *   37 → Tomato___healthy
+ *
+ * Some PlantVillage classes overlap with this app's knowledge base.
+ * Exact class name → disease ID mappings:
+ *   Potato___Early_blight     → "early-blight"
+ *   Potato___Late_blight      → "late-blight"
+ *   Squash___Powdery_mildew   → "squash-powdery-mildew"
+ *   Strawberry___Leaf_scorch  → "strawberry-leaf-scorch"
+ *   Tomato___Bacterial_spot   → "bacterial-leaf-spot-tomato"
+ *   Tomato___Early_blight     → "early-blight"
+ *   Tomato___Late_blight      → "late-blight"
+ *   Tomato___Septoria_leaf_spot → "septoria-leaf-spot"
+ * All other classes map to "unknown" and are filtered out during enrichment.
+ *
+ * After fine-tuning to the app's 93 disease classes, this file will be
+ * rewritten to match the new model's output layer.
  */
 
-import rawDiseases from "@/data/diseases.json";
+// ─── PlantVillage class names (in model output order) ────────────────────
-import type { Disease } from "@/lib/types";
 
-const diseases: Disease[] = rawDiseases as Disease[];
+const PLANTVILLAGE_CLASSES: string[] = [
+  "Apple___Apple_scab",
+  "Apple___Black_rot",
+  "Apple___Cedar_apple_rust",
+  "Apple___healthy",
+  "Blueberry___healthy",
+  "Cherry_(including_sour)___Powdery_mildew",
+  "Cherry_(including_sour)___healthy",
+  "Corn_(maize)___Cercospora_leaf_spot Gray_leaf_spot",
+  "Corn_(maize)___Common_rust_",
+  "Corn_(maize)___Northern_Leaf_Blight",
+  "Corn_(maize)___healthy",
+  "Grape___Black_rot",
+  "Grape___Esca_(Black_Measles)",
+  "Grape___Leaf_blight_(Isariopsis_Leaf_Spot)",
+  "Grape___healthy",
+  "Orange___Haunglongbing_(Citrus_greening)",
+  "Peach___Bacterial_spot",
+  "Peach___healthy",
+  "Pepper,_bell___Bacterial_spot",
+  "Pepper,_bell___healthy",
+  "Potato___Early_blight",
+  "Potato___Late_blight",
+  "Potato___healthy",
+  "Raspberry___healthy",
+  "Soybean___healthy",
+  "Squash___Powdery_mildew",
+  "Strawberry___Leaf_scorch",
+  "Strawberry___healthy",
+  "Tomato___Bacterial_spot",
+  "Tomato___Early_blight",
+  "Tomato___Late_blight",
+  "Tomato___Leaf_Mold",
+  "Tomato___Septoria_leaf_spot",
+  "Tomato___Spider_mites Two-spotted_spider_mite",
+  "Tomato___Target_Spot",
+  "Tomato___Tomato_Yellow_Leaf_Curl_Virus",
+  "Tomato___Tomato_mosaic_virus",
+  "Tomato___healthy",
+] as const;
 
-// ─── Constants ───────────────────────────────────────────────────────────────
+// ─── PlantVillage → App disease ID mapping ──────────────────────────────
-
-/** Index for the "healthy" class */
-export const HEALTHY_INDEX = 0;
-
-/** First index for actual disease classes */
-export const FIRST_DISEASE_INDEX = 1;
-
-/** Index for the "unknown" catch-all class */
-export const UNKNOWN_INDEX = 1 + diseases.length;
-
-/** Total number of output classes */
-export const NUM_CLASSES = UNKNOWN_INDEX + 1;
-
-// ─── Index → Disease ID mapping ──────────────────────────────────────────────
 
 /**
- * Map from model output index to disease ID string.
+ * Maps PlantVillage class names (in the form "Plant___Disease") to
- * Index 0 = "healthy", indices 1..N = disease IDs, last = "unknown".
+ * this app's disease IDs. Unmapped classes resolve to "unknown".
+ */
+function plantVillageNameToDiseaseId(pvName: string): string {
+  const parts = pvName.split("___");
+  if (parts.length !== 2) {
+    return "unknown";
+  }
+
+  const disease = parts[1];
+
+  // Detect "healthy" variants
+  if (disease === "healthy") {
+    return "healthy";
+  }
+
+  // Map exact PlantVillage class names to our disease IDs.
+  // Only map classes where we're confident the correspondence holds.
+  const exactMap: Record<string, string> = {
+    Squash___Powdery_mildew: "squash-powdery-mildew",
+    Strawberry___Leaf_scorch: "strawberry-leaf-scorch",
+    Potato___Early_blight: "early-blight",
+    Potato___Late_blight: "late-blight",
+    Tomato___Bacterial_spot: "bacterial-leaf-spot-tomato",
+    Tomato___Early_blight: "early-blight",
+    Tomato___Late_blight: "late-blight",
+    Tomato___Septoria_leaf_spot: "septoria-leaf-spot",
+  };
+
+  return exactMap[pvName] ?? "unknown";
+}
+
+// ─── Constants ──────────────────────────────────────────────────────────
+
+/** Total number of model output classes */
+export const NUM_CLASSES = PLANTVILLAGE_CLASSES.length; // 38
+
+/** Index for the "healthy" class — multiple PV indices map to this */
+export const HEALTHY_INDEX = 0; // First PV healthy class, others also map to this string
+
+/** First disease index (unused in PV mapping, kept for compatibility) */
+export const FIRST_DISEASE_INDEX = 0;
+
+/** Index for the "unknown" catch-all — PV classes we can't map */
+export const UNKNOWN_INDEX = NUM_CLASSES - 1; // 37 (Tomato___healthy maps to "healthy", not unknown)
+
+// ─── Index → Disease ID mapping ─────────────────────────────────────────
+
+/**
+ * Map from model output index to app disease ID string.
+ * Built dynamically from PlantVillage class names.
  */
 export const INDEX_TO_DISEASE_ID: Record<number, string> = Object.freeze(
   (() => {
     const map: Record<number, string> = {};
-    map[HEALTHY_INDEX] = "healthy";
+    for (let i = 0; i < NUM_CLASSES; i++) {
-    for (let i = 0; i < diseases.length; i++) {
+      map[i] = plantVillageNameToDiseaseId(PLANTVILLAGE_CLASSES[i]);
-      map[FIRST_DISEASE_INDEX + i] = diseases[i].id;
     }
-    map[UNKNOWN_INDEX] = "unknown";
     return map;
   })(),
 );
 
-// ─── Disease ID → Index mapping ──────────────────────────────────────────────
+// ─── Disease ID → Index mapping ─────────────────────────────────────────
 
 /**
  * Map from disease ID string to model output index.
+ * For duplicates (e.g., both potato and tomato "Early_blight" → "early-blight"),
+ * returns the first matching index.
  */
 export const DISEASE_ID_TO_INDEX: Record<string, number> = Object.freeze(
   (() => {
     const map: Record<string, number> = {};
-    map["healthy"] = HEALTHY_INDEX;
+    for (let i = 0; i < NUM_CLASSES; i++) {
-    for (let i = 0; i < diseases.length; i++) {
+      const id = INDEX_TO_DISEASE_ID[i];
-      map[diseases[i].id] = FIRST_DISEASE_INDEX + i;
+      // First occurrence wins (potato before tomato for early/late blight)
+      if (map[id] === undefined) {
+        map[id] = i;
+      }
     }
-    map["unknown"] = UNKNOWN_INDEX;
     return map;
   })(),
 );
 
-// ─── Lookup helpers ──────────────────────────────────────────────────────────
+// ─── Lookup helpers ─────────────────────────────────────────────────────
 
 /**
  * Get the disease ID for a given model output index.
@@ -93,9 +216,22 @@ export function isRealDisease(diseaseId: string): boolean {
   return diseaseId !== "healthy" && diseaseId !== "unknown";
 }
 
+/**
+ * Get the PlantVillage display name for a given model output index.
+ */
+export function getPlantVillageClassName(index: number): string {
+  return PLANTVILLAGE_CLASSES[index] ?? "unknown";
+}
+
 /**
  * Get all known disease IDs (excluding "healthy" and "unknown").
  */
 export function getAllDiseaseIds(): string[] {
-  return diseases.map((d) => d.id);
+  const ids = new Set<string>();
+  for (const id of Object.values(INDEX_TO_DISEASE_ID)) {
+    if (id !== "healthy" && id !== "unknown") {
+      ids.add(id);
+    }
+  }
+  return Array.from(ids);
 }

apps/web/src/lib/ml/model-loader.ts

@@ -93,7 +93,10 @@ export async function getModel(): Promise<PlantDiseaseModel> {
     const model = await Promise.race([
       loadingPromise,
       new Promise<never>((_, reject) =>
-        setTimeout(() => reject(new Error(`Model load timed out after ${MODEL_LOAD_TIMEOUT}ms`)), MODEL_LOAD_TIMEOUT),
+        setTimeout(
+          () => reject(new Error(`Model load timed out after ${MODEL_LOAD_TIMEOUT}ms`)),
+          MODEL_LOAD_TIMEOUT,
+        ),
       ),
     ]);
 
@@ -172,6 +175,18 @@ async function tryLoadTFJS(): Promise<PlantDiseaseModel | null> {
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
   let tf: any;
 
+  // Monkey-patch: add util.isNullOrUndefined for Node.js 26 compatibility.
+  // @tensorflow/tfjs-node references this function which was removed in Node 15+.
+  // eslint-disable-next-line @typescript-eslint/no-require-imports
+  const nodeUtil = require("util");
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  if (typeof (nodeUtil as any).isNullOrUndefined !== "function") {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    (nodeUtil as any).isNullOrUndefined = function (x: unknown): boolean {
+      return x === null || x === undefined;
+    };
+  }
+
   // Try tfjs-node first (server-side, uses native bindings).
   // Use dynamic strings so bundlers (Turbopack/webpack) don't trace these
   // as required dependencies — they are truly optional.
@@ -197,7 +212,9 @@ async function tryLoadTFJS(): Promise<PlantDiseaseModel | null> {
       const startTime = performance.now();
 
       // Reshape to [1, 3, 160, 160] NCHW → [1, 160, 160, 3] NHWC for TF.js
-      const inputTensor = tf.tensor4d(Array.from(tensor), [3, 160, 160])
+      // Reshape NCHW flat array [3*160*160] → [3, 160, 160] → NHWC [1, 160, 160, 3]
+      const inputTensor = tf
+        .tensor3d(Array.from(tensor), [3, 160, 160])
         .transpose([1, 2, 0])
         .expandDims(0);
 
@@ -352,7 +369,7 @@ function generateMockLogits(tensor: Float32Array): Float32Array {
   logits[topIndex] = 3.5;
 
   // Second highest
-  const secondIndex = (topIndex + Math.abs(hash % 10) + 1) % (numClasses - 1) + 1;
+  const secondIndex = ((topIndex + Math.abs(hash % 10) + 1) % (numClasses - 1)) + 1;
   logits[secondIndex] = 2.5;
 
   logits[numClasses - 1] = -2; // "unknown" gets low score