InsightHub
A document question answering platform that answers from your own files and cites the passage it used.
- 2026
- Personal project
- Retrieval augmented generation
Overview
InsightHub lets you upload documents and ask questions about them. The point of the project was the citation requirement: every answer has to name the passage it came from, so you can check it.
Ingestion runs documents through Tika, splits them into chunks, embeds them with AWS Bedrock and stores the vectors in PostgreSQL using pgvector. Search uses an HNSW index. The retrieved passages go to the Claude API, which answers only from what it was given.
That constraint made the project harder to build. It also made it usable, because a wrong answer that names its source is one you can catch.
What I built
- Built the ingestion pipeline from upload through Tika, chunking and embedding
- Set up pgvector with an HNSW index for semantic search
- Wrote the retrieval and prompting layer against the Claude API
- Built the React front end with streaming answers and inline citations
- Containerised the stack with Docker
E-Banking Platform