ML INFRASTRUCTURE · DISTRIBUTED TRAINING · LLM SYSTEMS · ON-DEVICE INFERENCE
I build the machinery around models — distributed training, on-device inference, agent tool infrastructure, and evaluation systems that catch hallucinations before users do.
M.S. Artificial Intelligence @ WPI '27 · building Zolli Labs
this profile runs like an ML system · 01 TRAIN → 02 EVAL → 03 SHIP → 04 MONITOR · fully automated by GitHub Actions
not an animation — a real model. a tiny policy is learning this level from visitor votes: it dies, you tell it what it should have done, and every sunday CI trains it and commits new weights.
one click below opens a ready-made ballot — just press submit. CI counts your vote in ~30s and the card updates.
every enemy block is one of my real commits — the ship re-fights this battle daily via GitHub Actions
|
⚡ Fault-tolerant distributed ML runtime — training jobs survive the machine being destroyed. Core of Zolli Labs.
|
📱 FLAN-T5 → ONNX on Android ARM64 — a QA transformer in your pocket, quantized without wrecking answers.
|
🧪 360-claim factorial study — grounding drops severe LLM fabrications from 34% to 0%.
|
|
🧠 Alzheimer's & depression classification from brain MRI scans with domain-adapted vision models.
|
🔩 A working GPT assembled from self-implemented parts — BPE, grouped-query attention, KV cache, training loops.
|
🚀 Turns GitHub repositories into STAR-method resume bullets and tailored cover letters. GoatHacks winner.
|
🔒 also in the lab: REU 2025 · Murai Lab — research in progress (repo private)
🏆 BENCHMARKS // HACKATHON RECORD
| Event | Build | |
|---|---|---|
| 🏆 | Runpod Flash Hackday '26 | FlashML — distributed ML engine, later grew into the Zolli Labs runtime |
| 🏆 | Berkeley AI Hackathon '26 | AI Hack Berkeley build |
| 🏆 | BetaFund × Stanford Hack | Cortex — team project (repo lives on a teammate's account) |
| 🏆 | HackResearch '26 | LLM Hallucination Study — started as a hackathon win |
| 🏆 | GoatHacks '25 | Repo2Resume — GitHub repos → STAR-method resume bullets |
| 🏆 | NASA Space Apps Challenge '25 | Space Apps build |
| ⚙️ | YC Hackathon '26 | YC_Hack |
| ⚙️ | Okta × Stripe Hack '26 | okta_stripe_hack |
| ⚙️ | Gen UI Hack '26 | Gen_UI_Hack |
| ⚙️ | Vortex Harness Engineering Hack '26 | Vortex-Harness |
| ⚙️ | AI for Education Hackathon '26 | AI-for-Education |
| ⚙️ | BrightData Hack '26 | BrightData |
🗄 MORE SYSTEMS
- SyntheSearch — AI research tool that finds and synthesizes the most relevant papers
- On_Device_Deep_Learning — deep learning benchmarks on phone hardware
- RecSys_MAG — recommender-system experiments on academic graph data
|
|
|
all telemetry regenerates itself daily — this profile is a running system, not a static page
