langchain-doc-summarizer

FastAPI pipeline for LangChain doc ingestion & summarization

---

1. Overview

langchain-doc-summarizer is a lightweight, production‑ready API that ingests documents (PDF, DOCX, TXT, etc.), processes them with LangChain’s powerful retrieval and embedding tools, and returns concise summaries. The service can be used as a microservice inside larger applications or run locally for quick experimentation.

---

2. Features

• FastAPI backend – Fast, async request handling with automatic OpenAPI docs.
• LangChain integration – Uses LangChain’s loaders, embeddings, and chain logic.
• File ingestion pipeline – Supports PDF, DOCX, TXT and CSV via app/utils/file_handler.py.
• Summarization service – Generates summaries using a pre‑trained LLM (OpenAI/Anthropic) through app/services/summarizer.py.
• Asynchronous task queue – Celery worker (tasks/celery_worker.py) for long‑running summarizations.
• Auth middleware – Simple token‑based authentication in middlewares/auth_middleware.py.
• CLI helper – cli/cli.py to seed data and run the server.
• Unit tests – Tests for document processing and summarization logic.

---

3. Tech Stack

Layer	Technology
API	FastAPI, Uvicorn
Background Jobs	Celery + Redis
LLM & Retrieval	LangChain (OpenAI / Anthropic)
Data Storage	Chroma DB (in‑memory by default)
Testing	Pytest
Packaging	Poetry

---

4. Installation

# Clone the repository
git clone https://github.com/yourorg/langchain-doc-summarizer.git
cd langchain-doc-summarizer

# Create virtual environment & install dependencies
poetry install

# Set required environment variables (example)
export OPENAI_API_KEY="sk-..."
export REDIS_URL="redis://localhost:6379"

---

5. Running the Service

# Start Redis locally or point to an existing instance
docker run -p 6379:6379 redis

# Launch Celery worker (in a separate terminal)
poetry run celery -A tasks.celery_worker worker --loglevel=info

# Run FastAPI server
poetry run uvicorn app.main:app --reload

Open the Swagger UI at http://localhost:8000/docs to try out endpoints.

---

6. Usage Examples

Upload a Document & Get Summary (synchronous)

curl -X POST "http://localhost:8000/api/v1/summarize" \
     -H "Authorization: Bearer YOUR_TOKEN" \
     -F "file=@/path/to/report.pdf"

Response:

{
  "summary": "The report highlights key findings about market trends..."
}

Upload a Document & Get Summary (asynchronous)

curl -X POST "http://localhost:8000/api/v1/summarize_async" \
     -H "Authorization: Bearer YOUR_TOKEN" \
     -F "file=@/path/to/report.pdf"

Response:

{
  "task_id": "123e4567-e89b-12d3-a456-426614174000",
  "status_url": "/api/v1/tasks/123e4567-e89b-12d3-a456-426614174000/status"
}

Poll the status endpoint until completed:

curl -X GET "http://localhost:8000/api/v1/tasks/123e4567-e89b-12d3-a456-426614174000/status" \
     -H "Authorization: Bearer YOUR_TOKEN"

---

7. API Endpoints

Method	Path	Description
POST	/api/v1/summarize	Upload file and receive immediate summary
POST	/api/v1/summarize_async	Upload file, returns task ID for background processing
GET	/api/v1/tasks/{task_id}/status	Get status of async summarization
DELETE	/api/v1/files/{file_id}	Remove stored file and embeddings

All endpoints require an Authorization: Bearer <token> header.

---

8. CLI Commands

# Seed sample documents into the vector store
poetry run python cli/cli.py seed

# Run the API server (alias for uvicorn)
poetry run python cli/cli.py serve

See cli/cli.py for additional options.

---

9. References and Resources

1. GitHub - bosukeme/document-summarizer-langgraph
2. Self Internet hosting RAG Purposes On Edge Units with Langchain
3. Building a RAG API with FastAPI – Analytics Vidhya
4. Building an AI-Powered Document QA API with FastAPI and LangChain
5. LangChain Overview – Docs by LangChain

---

10. Contributing

1. Fork the repository and create a feature branch (git checkout -b feature/<name>).
2. Run tests locally: poetry run pytest.
3. Submit a pull request with clear description of changes.

All contributions must follow PEP‑8 coding style and include unit tests where applicable.

---

11. License

MIT © 2026

---