A retrieval-grounded assistant that drafts triage notes for emergency nurses, checked against 40,000 past cases before anything reaches a chart.

An on-device vision model that counts warehouse stock from a handheld scanner — no network, no cloud round-trip, no waiting.

Multilingual semantic search across two million small-business listings, rebuilt from brittle keyword matching.

An open-source harness for regression-testing LLM features before every deploy, so prompt changes stop breaking production quietly.

Streaming speech-to-text tuned for noisy studio chatter, running under 200 ms end to end on a single consumer GPU.

Anomaly detection over satellite telemetry that flags failing parts days before an operator would notice.