Tapti Labs (@TaptiLabs) header
Tapti Labs (@TaptiLabs) profile picture

Understanding AI, ML, software & systems — through practical builds, workflows, and clear explanations. No hype.

Joined September 2026
Documents are visually complicated. Questions about them usually aren't. NanoVDR exploits that asymmetry: a big 2B vision model indexes your PDFs once, offline — but a tiny 69M-parameter, CPU-only text encoder handles every query afterward, at 51ms, retaining 95% accuracy.
1
8
1/2 You don't need a vision model to ask a question, only to read the page. That split is worth stealing for any retrieval system with expensive documents and cheap queries.
5
A useful rule in AI engineering: Don’t start with the model. Start with the failure mode. Ask: • What can go wrong? • Can rules solve part of it? • Where does ML add value? • How will we know it works? Good AI starts with good problem framing.
1
40