I'm Vinh, a full-stack and AI engineer in Atlanta. I'm happiest when I'm building, and the projects I care about most all started the same way: a real problem — a family searching for someone, a neighborhood that goes dark, a racer who needs a coach in their ear — and the question of what AI could actually do about it. I finished my B.S. in CS at Georgia State (Dean's List) and started the M.S. at Georgia Tech in August 2026. Before that I co-led a team at SkyIT modernizing a multi-tenant Django/React platform, and spent the summer as a CodePath Technical Fellow for a 200+ learner AI engineering course.
|
Now · M.S. CS @ Georgia Tech (OMSCS) · Shipping WATTLINE past the hackathon build · Local, on-device agents (Bearing Witness) |
Next · Agentic systems with a human in the loop · Multi-vector retrieval beyond forensics · Founding something in AI for high-stakes domains |
Hackathon record · 🥇 1st @ Actian VectorAI DB Build Challenge · 🏆 Best Hack for Good @ Hack RenderATL / MLH · ⚙️ Dell × NVIDIA AI Hackathon NYC · 🏎️ IBM SkillsBuild AI Builders · 🗺️ Team USA × Google Cloud · 🛡️ Reddit Mod Tools
|
🏆 WATTLINE Best Hack for Good · Hack RenderATL / MLH · Aug 2026 When the power goes out, some people are on a clock. WATTLINE maps electricity-dependent medical equipment users to neighborhood-level outage exposure across 25 Atlanta NPUs, then routes charging capacity to the highest-gap areas, constrained by MARTA reachability. I led the 4-person team: scoped the project, verified the federal HHS emPOWER pipeline against a 92,233-person state anchor, and built the React/MapLibre frontend with a 24-hour timeline scrubber. Read path is precomputed JSON behind FastAPI, no database per request. Python · FastAPI · React · MapLibre · HHS emPOWER · Live demo · Devpost |
|
🥇 Trace 1st Place · Actian VectorAI DB Build Challenge AI forensic search matching missing-person descriptions to unidentified remains. Multi-vector retrieval (SapBERT, BGE-M3, CLIP) fused with Reciprocal Rank Fusion. Python · FastAPI · React · Actian VectorAI Demo · DoraHacks |
⚙️ Bearing Witness Dell × NVIDIA AI Hackathon · NYC Always-on, fully local bearing-screening agent on the Dell Pro Max GB10. Physics does the diagnosis, a local model files the paperwork, a human has to say yes. Python · Local LLM · Signal processing · GB10 |
🏎️ APEX IBM SkillsBuild AI Builders Challenge AI race engineer coaching adaptive racers on IBM Granite. Owned the backend: 20 live endpoints, Granite tool integration, 192-test suite gating every merge. Python · FastAPI · IBM Granite · pytest · CI Demo |
|
🗺️ Hometown Pathway Atlas Team USA × Google Cloud Hackathon County-level Olympic and Paralympic parity analytics. Data pipeline with per-capita normalization and empirical-Bayes shrinkage; Vertex AI Gemini on Cloud Run. Python · Vertex AI Gemini · Cloud Run Demo |
🛡️ ContextMod Reddit Mod Tools Hackathon Rule-engine backend for a native Devvit moderation bot: JSON5 rules, multiple rule kinds and action handlers, strict TypeScript, fully tested. TypeScript · Devvit · JSON5 App |
🎮 Rivals Coach Side project Live coaching companion for Marvel Rivals: reads the match roster via Overwolf GEP and suggests one pick / swap / hold from your own comfort pool. Decision support, not a stats tracker. TypeScript · Overwolf |
More on Devpost · vinh-le.com
|
Tests gate the merge APEX shipped with a 192-test backend suite wired into CI before the demo, not after. If it isn't tested, it isn't done. |
Precompute the read path WATTLINE serves frozen JSON; nothing calculates per request. Boring at runtime is a feature. |
A human says yes Bearing Witness drafts the inspection; it never files it. AI proposes, a person decides, especially when the stakes are real. |
| LANGUAGES |
|
| AI / ML |
|
| BACKEND |
|
| FRONTEND |
|
| DATA |
|
| CLOUD / TOOLS |
|
Built by Vinh Le · Atlanta, GA
Last edited: August 2026

