AI Systems Engineer

Hi, I'm Pradhumn.

I build the whole stack of an AI product. The retrieval, the systems underneath it, and the interface on top.

Gallery

14 things that should not run in a browser

Selected work

2026

Search that answers what to eat, not just where. Built end to end.

2026

Control your own machine from any browser. 60fps H.264 over QUIC, with no relay server.

60fps screen streaming · 7.4 KB per frame, down from 267 · 0 relay servers

2026

GPGPU flow fields, morphing particles, ocean and galaxy simulation. Live, not screenshots.

14 pieces, all deployed

Interactive · The RAG Challenge

Balance accuracy against latency and cost across three levels. It is the trade every real retrieval system lives or dies by, made playable.

Handle zero results in agentic RAG search by relaxing hard payload filters progressively, then falling back to a broad semantic query.

Tables break RAG because flattening them destroys row/column relationships; keep them as HTML and attach headers to every row to fix it.

When RAG breaks, the model usually isn't the problem — retrieval is, about 73% of the time. Here's the real failure map and what to fix first.

Qdrant applies payload filters during HNSW graph traversal, not after — so filtered queries stay fast and always return a full top-k.

Occasional field notes on production RAG and LLM systems. No spam, unsubscribe anytime.