Halioo

Halioo is a full-stack Retrieval-Augmented Generation chatbot. Users upload PDF or TXT documents and ask questions in natural language; answers are generated from the content of those documents and every answer cites its sources. Responses stream token by token over Server-Sent Events, each user's documents are isolated behind JWT authentication, and chat sessions persist across visits. The service runs live with the frontend on Vercel, the backend on Render, and the database on Neon.

General-purpose chatbots can't answer questions about private documents, and when pushed they hallucinate — for document-heavy work, an answer without a verifiable source is worthless.

A RAG pipeline: uploaded documents are chunked and embedded into PostgreSQL with pgvector, questions retrieve the most similar chunks via vector similarity search, and Google Gemini generates answers constrained to that retrieved context — with citations pointing back to the source passages. SSE streaming keeps responses immediate, and per-user data isolation keeps documents private.

I wanted to understand RAG beyond API-wrapper tutorials — embeddings, vector search, retrieval quality, and streaming — by building and operating the whole pipeline myself as a live production service.

• React
• Node.js
• Express
• PostgreSQL
• pgvector
• Prisma
• Google Gemini API
• LangChain

Tried this project or read the case study? I'd love to hear your thoughts.