KnowledgeNav.io lets you upload documents (PDF, DOCX, TXT, HTML), ask questions in a chat interface, and receive answers that are strictly grounded in your documents — every factual claim is backed by a [SOURCE] citation that links to the exact chunk it came from. Hallucinations are detected and blocked before they reach the user.
- Hybrid retrieval — BM25 keyword search + Cohere multilingual vector search, fused with Reciprocal Rank Fusion (RRF)
- Cross-encoder reranking — Cohere rerank-v3.5 scores the top candidates before generation
- Strict citation enforcement — LLM must cite every factual sentence; the pipeline validates coverage and rejects hallucinated chunk IDs, retrying if needed
- Agentic workflow — LangGraph graph with analyse → retrieve → rerank → generate → validate → persist nodes
- Streaming responses — Server-Sent Events deliver tokens as they arrive
- Multi-corpus — Create isolated corpora per project; documents are scoped to their corpus
- OAuth authentication — Google and GitHub login, JWT access + refresh token rotation
- RAGAS evaluation — Built-in evaluation pipeline measuring faithfulness, answer relevancy, context precision, and context recall
| Layer | Technology |
|---|---|
| Frontend | React 18, TypeScript, Vite, Tailwind CSS 4 |
| Backend | FastAPI, Python 3.11, LangGraph, LangChain Core |
| LLM | Groq (llama-3.3-70b-versatile) |
| Embeddings | Cohere embed-v3-multilingual |
| Reranking | Cohere rerank-v3.5 |
| Vector DB | ChromaDB |
| Keyword search | rank-bm25 |
| Relational DB | PostgreSQL 16 (asyncpg + SQLAlchemy 2) |
| Cache / rate-limit | Redis 7 |
| Migrations | Alembic |
| Auth | python-jose (JWT), Google OAuth2, GitHub OAuth |
| Evaluation | RAGAS |
| Observability | OpenTelemetry, LangSmith |
| CI/CD | GitHub Actions, Railway (API), Vercel (frontend) |
- Python 3.11+
- Node.js 20+
- Docker + Docker Compose
git clone https://github.com//Leela0o5/KnowledgeNav.io
cd KnowledgeNav.io
# Copy and fill in environment variables
cp .env.example .env
# Start infrastructure (PostgreSQL, Redis, ChromaDB)
docker compose up -d postgres redis chroma
# Install backend
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -e ".[dev]"
# Run database migrations
alembic upgrade head
# Start API
uvicorn api.main:app --reload --host 0.0.0.0 --port 8080
# Install and start frontend (separate terminal)
cd frontend
npm install
npm run devThe app will be available at http://localhost:3000. The API runs on http://localhost:8080.
docker compose upThis starts the API, frontend, PostgreSQL, Redis, ChromaDB, and OpenTelemetry collector.
Copy .env.example to .env and fill in the required values.
See .env.example for all available settings and their defaults.
- Sign in with Google or GitHub on the landing page
- Create a corpus — click "New Corpus" in the sidebar
- Upload documents — click the upload icon and add PDF, DOCX, TXT, or HTML files
- Ask questions — type in the chat input; responses stream with inline
[SOURCE]citations - Review citations — click any citation to see the exact source chunk
- Switch corpora — select a different corpus from the sidebar to query different document sets
- Manage sessions — previous conversations are saved per corpus and accessible from the sidebar
- Fork the repository and create a feature branch
- Install dev dependencies:
pip install -e ".[dev]" - Run linting:
ruff check backend - Run type checking:
mypy backend - Run tests:
pytest - Open a pull request against
main
CI will run lint, type-check, unit tests, integration tests, and a frontend type-check on every PR.