fix(vision): enable ML training on production server

Research confirmed TensorFlow has wheels for Linux x86_64 (including
Synology NAS). The previous failures were due to:
1. Training scripts not being copied to Docker image
2. Venv path not being writable in container

Changes:
- Update isPlatformSupported() to include Linux ARM64 support
- Move venv to data/vision-training/.venv (mounted volume, writable)
- Add python3-venv to Docker image dependencies
- Copy training scripts to Docker image
- Create vision-training directories in Docker build

Sources:
- https://www.tensorflow.org/install/pip
- https://pypi.org/project/tensorflow/ (shows manylinux_2_17_aarch64 wheels)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

.claude/settings.local.json

@@ -568,7 +568,8 @@
       "Bash(scripts/train-column-classifier/.venv/bin/python:*)",
       "Bash(brew list:*)",
       "Bash(scripts/train-column-classifier/.venv/bin/pip index:*)",
-      "Bash(scripts/train-column-classifier/.venv/bin/pip:*)"
+      "Bash(scripts/train-column-classifier/.venv/bin/pip:*)",
+      "WebFetch(domain:pypi.org)"
     ],
     "deny": [],
     "ask": []

Dockerfile

@@ -110,9 +110,11 @@ FROM node:20-slim AS runner
 WORKDIR /app
 
 # Install ONLY runtime dependencies (no build tools)
+# python3-venv is needed for creating virtual environments for ML training
 RUN apt-get update && apt-get install -y --no-install-recommends \
     python3 \
     python3-pip \
+    python3-venv \
     qpdf \
     ca-certificates \
     && rm -rf /var/lib/apt/lists/*
@@ -154,6 +156,9 @@ COPY --from=builder --chown=nextjs:nodejs /app/packages/templates ./packages/tem
 # Copy abacus-react package (needed for calendar generation scripts)
 COPY --from=builder --chown=nextjs:nodejs /app/packages/abacus-react ./packages/abacus-react
 
+# Copy ML training scripts (for vision model training)
+COPY --from=builder --chown=nextjs:nodejs /app/apps/web/scripts/train-column-classifier ./apps/web/scripts/train-column-classifier
+
 # Install Python dependencies for flashcard generation
 RUN pip3 install --no-cache-dir --break-system-packages -r packages/core/requirements.txt
 
@@ -164,8 +169,8 @@ COPY --from=builder --chown=nextjs:nodejs /app/apps/web/package.json ./apps/web/
 # Set up environment
 WORKDIR /app/apps/web
 
-# Create data directory for SQLite database and uploads
+# Create data directory for SQLite database, uploads, and vision training
-RUN mkdir -p data/uploads && chown -R nextjs:nodejs data
+RUN mkdir -p data/uploads data/vision-training/collected data/vision-training/.venv && chown -R nextjs:nodejs data
 
 USER nextjs
 EXPOSE 3000

apps/web/.gitignore

@@ -64,4 +64,5 @@ data/uploads/
 # ML training data
 training-data/
 data/vision-training/collected/
+data/vision-training/.venv/
 scripts/train-column-classifier/.venv/

apps/web/src/app/api/vision-training/config.ts

@@ -17,30 +17,27 @@ const cwd = process.cwd()
 
 /**
  * Check if the current platform supports TensorFlow training.
- * TensorFlow doesn't have wheels for all platforms (e.g., ARM-based NAS devices).
+ * TensorFlow has wheels for:
+ * - macOS x86_64 and arm64 (Apple Silicon with tensorflow-macos)
+ * - Linux x86_64 and aarch64
+ * - Windows x86_64
  */
 export function isPlatformSupported(): { supported: boolean; reason?: string } {
   const platform = process.platform
   const arch = process.arch
 
-  // TensorFlow supports:
-  // - macOS on x86_64 and arm64 (Apple Silicon with tensorflow-macos)
-  // - Linux on x86_64 (and some arm64 builds)
-  // - Windows on x86_64
-
   if (platform === 'darwin') {
     // macOS - both Intel and Apple Silicon are supported
     return { supported: true }
   }
 
   if (platform === 'linux') {
-    if (arch === 'x64') {
+    if (arch === 'x64' || arch === 'arm64') {
       return { supported: true }
     }
-    // ARM Linux (like Synology NAS) typically doesn't have TensorFlow wheels
     return {
       supported: false,
-      reason: `TensorFlow is not available for Linux ${arch}. Training should be done on a machine with x86_64 or Apple Silicon.`,
+      reason: `TensorFlow is not available for Linux ${arch}. Training requires x86_64 or ARM64.`,
     }
   }
 
@@ -50,7 +47,7 @@ export function isPlatformSupported(): { supported: boolean; reason?: string } {
 
   return {
     supported: false,
-    reason: `TensorFlow is not available for ${platform} ${arch}. Training should be done on macOS, Linux x86_64, or Windows x86_64.`,
+    reason: `TensorFlow is not available for ${platform} ${arch}. Training should be done on macOS, Linux (x86_64/ARM64), or Windows x86_64.`,
   }
 }
 
@@ -61,8 +58,12 @@ export const TRAINING_SCRIPTS_DIR = path.join(cwd, 'scripts/train-column-classif
 
 /**
  * Path to the venv directory
+ * We use data/vision-training/.venv because:
+ * 1. The data/ directory is mounted as a volume in Docker (persists across restarts)
+ * 2. It's writable by the container
+ * 3. Scripts directory may not exist in production Docker images
  */
-const VENV_DIR = path.join(TRAINING_SCRIPTS_DIR, '.venv')
+const VENV_DIR = path.join(cwd, 'data/vision-training/.venv')
 
 /**
  * Path to the Python executable in the venv