PGPradhumn Gupta

Interface · 2026 · Shipped

The RAG Challenge

Tune a production retrieval pipeline against accuracy, latency and cost. Playable.

levels
3levels

My part
Everything

Live ↗

Every retrieval system is a trade between how right the answer is, how long it takes, and how much it costs. Reading that sentence teaches nobody anything. Having to make the trade yourself, and watching one number get worse every time you improve another, teaches it in about two minutes.

So it is a game. Three levels, each one a pipeline you tune. The scoring uses the same shape of trade-off a real system faces.

It is also the most-visited thing on this site that is not a blog post, which told me something about what people actually want from a technical site.

← All work