I keep picking problems a little out of my depth🪨then poking at them till they give in🧩,betting one of these swings moves the world forward🚀 (probably)

A production knowledge & reasoning layer for AI agents. Documents become temporal knowledge graphs that answers can be traced back to.

95.1%

MultiHop-RAG accuracy

+28%

over HippoRAG-2 on MuSiQue

sub-second

graph retrieval

Diagnosing and training the memory-update gap in LLM agents: the first RL environment whose reward targets temporal fact-currency.

9.0% → 16.7%

held-out accuracy (+86%)

92% → 77%

frontier gap, bounded memory

An open-source RAG library with an adaptive per-query router that hits SOTA-tier accuracy at ~40% lower median latency.

F1 0.54 / EM 0.40

MuSiQue (gpt-4o-mini)

−40%

median latency

Field SOTA

matched, measured

A web app that reads a chest X-ray and returns a structured, radiologist-style report, built on Stanford's CheXagent and served on a scale-to-zero GPU so it costs nothing at rest.

A Random Forest that predicts corporate credit ratings from financial ratios at ~79% accuracy, trained on 7,805 real agency ratings and running entirely in the browser.