Grounded RAG & Knowledge-Graph API

A FastAPI service for grounded question answering, vector similarity search, and lightweight knowledge-graph extraction/querying.

This repository consolidates and updates related work for easier maintenance and reference.

Capabilities

• Grounded answering — answer from supplied context chunks with citations, confidence, and answerability metadata
• Vector search — cosine-similarity ranking
• Knowledge-graph extraction — entities and relations from source text
• Graph querying — structured relationship queries
• Community summaries — higher-level graph summaries
• Input validation — Pydantic request/response models
• Health endpoint — deployment-friendly service check

API surface

Method	Endpoint	Purpose
POST	/grounded-answer	Context-grounded Q&A with citations
POST	/vector-search	Similarity-based retrieval
POST	/extract-graph	Entity/relation graph extraction
POST	/graph-query	Structured graph queries
POST	/community-summary	Community-level graph summaries
GET	/health	Service health check

Run locally

python -m venv .venv
source .venv/bin/activate
pip install -r requirements-dev.txt
uvicorn main:app --reload

Open http://127.0.0.1:8000/docs for the interactive API.

Tests

pytest -q

Design

Several retrieval and scoring steps are kept deterministic so behaviour is easy to inspect and test.

Deployment

uvicorn main:app --host 0.0.0.0 --port $PORT

Use /health as the service health-check path.