Chinese OCR 2

- Add entrypoint script to auto-download OCR models on first run
- Include ch_sim (Simplified) and ch_tra (Traditional) for HK law firm use
- Add docling_models volume for model persistence
- Bump start_period to 120s for first-run download time

Code/TODO.md

@@ -23,18 +23,50 @@
 
 Rationale: De-risk unknowns (Docling OCR quality, Ollama CPU inference speed) before building orchestrator.
 
-### 1a. Docker Compose + Docling
+### 1a. Docker Compose + Docling (DONE)
 
-- [ ] Docker Compose base (Docling service)
+- [x] Docker Compose base (Docling service)
-- [ ] Test Docling API with sample PDFs (curl)
+- [x] Test Docling API with sample PDFs (curl)
-- [ ] Validate OCR quality on real PDFs + generated samples
+- [x] Validate OCR quality on real PDFs + generated samples
+- [x] Test OCR on images (PNG) — works with `force_ocr=true`
 
-### 1b. Ollama
+**Docling API Spec:**
 
-- [ ] Add Ollama to Docker Compose (Qwen3-1.7B, CPU mode)
+- Image: `ds4sd/docling-serve:latest`
-- [ ] Test structured extraction prompt (curl)
+- Port: `5001`
-- [ ] Measure inference time per document
+- Endpoint: `POST /v1/convert/file`
-- [ ] Validate extraction accuracy (deadline, doc type, etc.)
+- Form params: `files=@<path>;type=application/pdf`, `to_formats=md`, `do_ocr=true`
+- Response: `{ "document": { "md_content": "..." }, "status": "success" }`
+- OCR engine: `easyocr` (default), supports CJK
+
+**OCR Options:**
+
+- `do_ocr=true` — enabled by default, processes bitmap content
+- `force_ocr=true` — use for images/scanned PDFs (replaces existing text with OCR)
+- `ocr_engine` — `easyocr` (default), `tesseract`, `rapidocr`, `tesserocr`, `ocrmac`
+- `ocr_lang` — language codes (engine-specific), e.g. `ch_sim` for Simplified Chinese
+
+**Tested Formats:**
+
+- PDF (Filing Receipt) — extracted text, tables, CJK content ✅
+- PNG (Email screenshot) — extracted subject, dates, recipient ✅ (CJK trademark garbled, may need `ocr_lang`)
+
+### 1b. Ollama (DONE)
+
+- [x] Add Ollama to Docker Compose (Qwen3-1.7B, CPU mode)
+- [x] Test structured extraction prompt (curl)
+- [x] Measure inference time per document (~15-30s on 4 vCPU)
+- [x] Validate extraction accuracy (deadline, doc type, etc.)
+
+**Ollama API Spec:**
+
+- Image: `ollama/ollama:0.9.3`
+- Port: `11434`
+- Model: `qwen3:1.7b` (~1.4 GB, auto-pulled on startup)
+- Endpoint: `POST /api/generate`
+- Body: `{"model":"qwen3:1.7b","prompt":"...","stream":false,"options":{"temperature":0}}`
+- Response: `{ "response": "<json>" }` (strip `<think>...</think>` tags in post-processing)
+- Note: Qwen3 includes reasoning by default; `/no_think` leaves empty tags, so strip instead
 
 ### 1c. Python Tools Service
 

Code/docker-compose.yaml

@@ -3,20 +3,34 @@ services:
     docling:
         # https://github.com/docling-project/docling-serve
         # 5001: Web UI + API
+        # OCR: EasyOCR with English + Chinese Simplified + Traditional (auto-downloaded on first run)
 
         image: ghcr.io/docling-project/docling-serve-cpu:v1.13.0
         container_name: jt-docling
         restart: unless-stopped
         environment:
             DOCLING_SERVE_ENABLE_UI: "true"
+        volumes:
+            - docling_models:/opt/app-root/src/.cache/docling/models
         ports:
             - "5001:5001"
+        entrypoint: ["/bin/sh", "-c"]
+        command:
+            - |
+              # Download Chinese OCR models if missing (first run only)
+              if [ ! -f /opt/app-root/src/.cache/docling/models/EasyOcr/zh_sim_g2.pth ] || [ ! -f /opt/app-root/src/.cache/docling/models/EasyOcr/zh_tra_g2.pth ]; then
+                echo "Downloading Chinese OCR models (Simplified + Traditional)..."
+                python3 -c "import easyocr; easyocr.Reader(['en','ch_sim','ch_tra'], gpu=False)"
+                # Copy downloaded models to correct location
+                cp -n /opt/app-root/src/.EasyOCR/model/*.pth /opt/app-root/src/.cache/docling/models/EasyOcr/ 2>/dev/null || true
+              fi
+              exec uvicorn docling_serve.app:app --host 0.0.0.0 --port 5001
         healthcheck:
             test: ["CMD", "curl", "-sf", "http://localhost:5001/health"]
             interval: 30s
             timeout: 10s
             retries: 3
-            start_period: 60s
+            start_period: 120s
 
     # Ollama
     ollama:
@@ -55,5 +69,7 @@ services:
     #         retries: 3
 
 volumes:
+    docling_models:
+        name: jt-docling-models
     ollama_data:
         name: jt-ollama-data