← All work AI · Data · 2025

AI knowledge assistant

Confidential — SaaS

A RAG-powered assistant that answers from a company’s private docs with citations and guardrails.

LLMRAGPythonpgvector
−55% Resolution time
94% Answer accuracy
38k Docs indexed

The challenge

Support teams spent hours searching scattered internal documentation to answer customer questions.

What we did

We built a retrieval pipeline over pgvector, integrated Claude with citation enforcement, and embedded it into their helpdesk.

Tech stack

PythonClaudepgvectorFastAPISvelteKit