OCR Pipeline

PDF → OCR → NER → Summarization → TTS (audio)

Architecture

PDF → [Nginx] → [API Gateway] → [CPU Worker: render + postprocess]
                                      ↓
                               [GPU Worker: OCR → Summarize → TTS]
                                      ↓
                                 [MinIO: artifacts]
                                      ↓
                                 [Audio WAV]

Requirements

• Docker & Docker Compose
• NVIDIA GPU (RTX 3060+)
• NVIDIA Container Toolkit
• 16GB+ RAM

Quick Start

# 1. Clone and configure
cp .env.example .env
# Edit .env with your settings (HF_TOKEN, API_KEYS, etc.)

# 2. Start all services
make up

# 3. Upload a PDF
curl -X POST -F "file=@document.pdf" http://localhost:8000/pdf

# 4. Check job status
curl http://localhost:8000/jobs/<job_id>

# 5. Health check
curl http://localhost:8000/health

Services

Service	Role	GPU	Port
nginx	Reverse proxy	❌	80/443
redis	Job queue (RQ)	❌	6379
minio	Artifact storage	❌	9000/9001
api-gateway	FastAPI + upload endpoint	❌	8000
cpu-worker	PDF render + NER postprocess	❌	—
gpu-worker	OCR → Summarize → TTS	✅	—
ollama	Llama 3.1 8B for summarization	✅	11434

Pipeline

1. Upload — PDF → job created → enqueue render
2. Render — PyMuPDF → PNG pages → MinIO → enqueue OCR
3. OCR — PaddleOCR (GPU, Russian) → JSON → MinIO → enqueue postprocess
4. Postprocess — Natasha NER + regex cleanup → entities → MinIO → enqueue summarize
5. Summarize — Ollama + Llama 3.1 8B → summary text → MinIO → enqueue TTS
6. Translate — EN → RU (if summary contains English)
7. TTS — Silero v5_ru → WAV audio → MinIO → complete

Deployment

1. Configure Environment

cp .env.example .env

Edit .env:

# Required
HF_TOKEN=hf_your_token_here
API_KEYS=your-secret-key-1,your-secret-key-2

# Optional (defaults work for local dev)
REDIS_URL=redis://redis:6379/0
MINIO_USER=minioadmin
MINIO_PASSWORD=minioadmin
STORAGE_BACKEND=minio
LOG_FORMAT=console  # or "json" for production

2. Generate SSL Certificate (Production)

make ssl-gen

Or use Let's Encrypt:

certbot certonly --standalone -d your-domain.com
cp /etc/letsencrypt/live/your-domain.com/fullchain.pem ssl/cert.pem
cp /etc/letsencrypt/live/your-domain.com/privkey.pem ssl/key.pem

3. Start Services

make up

4. Verify

# Health check
curl http://localhost/health

# Upload test
curl -X POST -F "file=@test.pdf" http://localhost/pdf \
  -H "X-API-Key: your-secret-key-1"

# Check status
curl http://localhost/jobs/<job_id>

5. Monitor

make logs           # All services
make logs | grep gpu-worker  # GPU worker only

Development

make test       # Run tests
make lint       # Run ruff
make logs       # Tail logs
make shell-api  # Shell into api-gateway
make shell-gpu  # Shell into gpu-worker

Project Structure

ocr-pipeline/
├── shared/                  # Shared code (config, schemas, storage)
├── services/
│   ├── api_gateway/         # FastAPI gateway + middleware
│   ├── cpu_worker/          # Render + postprocess
│   └── gpu_worker/          # OCR + summarize + TTS
├── tests/                   # Unit + integration tests
├── nginx.conf               # Reverse proxy config
├── docker-compose.yml
├── Makefile
├── .env                     # Secrets (not in git)
└── .env.example             # Template

API Reference

Method	Endpoint	Auth	Description
POST	/pdf	API Key	Upload PDF, start pipeline
GET	/jobs/{id}	API Key	Get job status + result
GET	/health	No	Health check
GET	/docs	No	Swagger UI

License

MIT