AI Research Assistant
RAG-powered assistant that answers questions over large document corpora with cited sources.
Role: Solo developerDate: March 2025
- Next.js
- Python
- FastAPI
- LangChain
- pgvector
- OpenAI
Overview
A retrieval-augmented generation (RAG) system that lets users ask natural-language questions across thousands of documents and get answers grounded in cited sources. Built to minimize hallucination by always surfacing the underlying passages.
Key Features
- Hybrid semantic + keyword retrieval over a pgvector index.
- Streaming responses with inline source citations.
- Incremental ingestion pipeline that chunks, embeds and indexes new documents.
Architecture
A FastAPI backend handles ingestion and retrieval, Postgres with pgvector stores embeddings, and a Next.js frontend streams tokens to the client over server-sent events.