I work at the intersection of applied AI and healthcare: multi-agent systems, RAG, and LLM tooling that ships. On the side, I run vibewithsnehal — turning dense AI concepts into plain English for a general audience.
| Project | What it does |
|---|---|
Code Dot  |
One-press code coach for Cursor and Stream Deck: reviews the selected file with repository context, explains its purpose and logic, and speaks bugs, suggested fixes, and alternative approaches aloud. Powered by Codex, with support for unsaved edits. |
| CourtVision |
AI tennis line calling and match stats, live or from video — finds the court via homography, tracks the ball, detects bounces from trajectory physics, and calls IN/OUT with margins + confidence ~0.2s after the ball lands, streaming an annotated feed to the browser. Rallies, shot speeds, and bounce heatmaps collected just by watching. 100% local OpenCV, verified against a physics-simulated ground truth (9/9 calls correct, live = batch parity). |
AgentBoard  |
Local-first Jira alternative where AI agents are teammates — LangGraph + LangChain on local Ollama models work issues live on the board (solo & swarm), write real files, and push cards to Done with a peer-review sweep. Includes an opt-in Claude Code board that tracks and launches headless sessions with human-in-the-loop permissions. |
| face-mood-tracker |
Real-time face tracking that predicts mood from facial micro-expressions — MediaPipe blendshapes + FACS rules with smoothing/hysteresis, 100% local, SQLite capture and pattern analytics (timelines, heatmaps, transition matrices). |
| 508-checker |
Section 508 / WCAG accessibility scanner for PDF, Office docs, images & HTML — 100% local, with AI review via a self-hosted Ollama model. |
| syntra |
Design agent built at the TED AI Hackathon. |
| MedicaidAIGovHub |
Governance hub mapping a NIST overlay onto Medicaid. |
| afib-patient-level-screening |
Patient-level atrial fibrillation screening from wearable single-lead ECG — real, non-extrapolated false-positive rates. |
| medical-device-complaints |
Medical-device complaint pipeline with multi-agent AI processing, FDA triage, and AWS AgentCore deployment. |
| hutch-connector |
Matches a capabilities statement to ranked SAM.gov opportunities, with a live agentic (LangChain) bid checklist. |
Local mission control for AI agents: assign an issue, watch agents ship it on the board.
What I'm focused on
🤖 Multi-agent systems & LLM tooling that ships · 🩺 Applied AI in healthcare · 📣 Making AI make sense @vibewithsnehal
“AI, made human.”