Year2024
RoleFull-stack Engineer
Duration3 months
ClientOpen-source
StackNext.js, OpenAI API, Pinecone, TypeScript
02

The challenge

Knowledge-heavy teams spent hours hunting through PDFs and wikis. Off-the-shelf chatbots hallucinated and lacked source attribution.

The approach

Built a RAG pipeline with vector embeddings and citation-first prompts. Streaming responses, source links, and a focused chat UI tuned for trust and verification.

The outcome

  • 95% factual accuracy on internal benchmark
  • Average answer time: 2.4 seconds
  • Saves 6+ hours per researcher per week

Like what you see?

Have a project I could help with? Let's talk.

Get in touch