🤖 AI / MLIntermediate

Building a RAG Pipeline: From Chaos to 90%+ Recall

I rebuilt the same RAG system five times. Chunking strategy, embedding choice, and reranking moved my eval recall from 41% to 93% — here is each change, measured.

Sandesh Bramhane@sandesh3 min read1344 views

Spoiler-sensitive walkthrough

This writeup reveals a complete solution. If you're attempting it yourself, close this page — and come back when you're stuck or done.

your choice is remembered on this device

written by

Sandesh Bramhane

Full-stack developer & cybersecurity enthusiast. I build things, break things, and document the whole journey — from TryHackMe rooms to hackathon projects.

ReactNode.jsTypeScriptC++SQL
View profile →

Continue the thread

Discussion (1)

Sign in or create an account to join the discussion.

  • Did you measure token cost per query after the reranking step? Curious how much it added.