KnowFlow

Internal Knowledge Assistant powered by RAG + LangChain
Upload documents, ask natural language questions, and get contextual answers instantly.

Overview

KnowFlow is a Retrieval-Augmented Generation (RAG) system that provides contextual answers to user queries based on internal documents.
It uses FastAPI as the backend, vectorizes documents into a FAISS index, and routes tasks to appropriate agents using multi-agent logic.

Tech Stack

• FastAPI – Web API server
• LangChain – Chain and document retrieval logic
• FAISS – Local vector database
• OpenAI GPT-3.5 / GPT-4 – Answer generation
• HuggingFace Embedding – snunlp/KR-SBERT-V40K-klueNLI-augSTS
• Sentence Transformers – Used for multi-prompt embedding similarity-based routing
• Pydantic + pydantic-settings – Configuration management
• Uvicorn – ASGI server
• Streamlit – Lightweight UI demo

File Descriptions

• main.py – FastAPI application entry point
• config.py – Loads environment variables
• router.py – Multi-agent routing logic
• agents/ – Includes SummaryAgent, QAAgent, NERAgent, ClassificationAgent, and FallbackAgent
• retriever.py – FAISS-based document retriever
• generator.py – GPT-based response generator
• ingestion.py – Loads, splits, embeds documents and saves vectorstore
• streamlit_app.py – Streamlit-based UI demo
• utils.py – Utility functions (stub)
• agent.py – Agent interface definition (stub)
• data/docs/ – Directory for raw text and PDF files
• data/vectorstore/ – FAISS vector index storage

Features

• RAG-based response generation
• Named entity extraction, classification, summarization
• REST API endpoint (/ask)
• Full Korean embedding support

Usage

1. Set up .env

OPENAI_API_KEY=your-openai-key
EMBEDDING_MODEL=snunlp/KR-SBERT-V40K-klueNLI-augSTS 
VECTORSTORE_PATH=./data/vectorstore/index.faiss

2. Install dependencies

pip install -r app/requirements.txt

3. Embed documents

PYTHONPATH=. python core/ingestion.py

4. Run the API server

uvicorn app.main:app --reload
→ 브라우저에서 http://localhost:8000/docs 접속

5. Launch the Streamlit UI

streamlit run interface/streamlit_app.py

TODO

• Enhance multi-agent routing with LangGraph or FSM
• Upload API for new documents
• Automatic report generation (PDF/Markdown)
• Save chat history per user
• Support for local LLMs (e.g., Ollama, GPT4All)

License

MIT License © 2025 KnowFlow