About
I'm a Mechanical Engineering PhD student at Georgia Tech's MiNDS Lab, building ML-based control frameworks for two-phase cooling systems — electronics, EVs, high-energy lasers, nuclear. Before grad school, I spent two years as an R&D engineer at Advanced Cooling Technologies working on DOE SBIR, ARPA-E projects, and patent filings at TRL 8. On the side, I build LLM vision pipelines and AI-native apps.
Work Experience
Skills
Projects
LLM vision pipelines, AI-native apps, and ML systems — built end-to-end.
Point at any location on the globe, pull live Sentinel-2 imagery, and get instant AI analysis — all in ~5 seconds. Built a full-stack pipeline: interactive web globe → Sentinel-2 STAC API → false-color spectral composites (NDRE/NDVI/NDMI encoded as RGB) → fine-tuned LFM2.5-VL that produces structured triage JSON (stressor type, severity, ANOMALOUS/NORMAL). LoRA fine-tuned at 1.69% trainable params on 443 Iowa corn belt tiles with USDA yield cross-validation (r=−0.505, p<0.0001). 50,000:1 compression vs. raw imagery downlink. Built for the Liquid AI × DPhi "AI in Space" Hackathon.
GIN ensemble pipeline for robust drug-target binding prediction across KIBA (CI=0.882), Davis (CI=0.735), and BindingDB (CI=0.625, AUROC=0.760) — with uncertainty quantification for abstention on low-confidence predictions. Multi-agent RAG architecture for biomedical reasoning and decision support. Full reproducibility package: code, experiments, LaTeX paper, poster, training logs. Built for the GT STAR-AI Makerspace Hackathon (May 2026, Petit Institute for Bioengineering). Trained on Apple Silicon M4.
Tell it what's in your pantry. Get a recipe. Smart pantry ranker uses nomic-embed-text embeddings + section dampening + usage frequency to surface the best ingredients first. Local LLM (llama3.2:3b via Ollama over Tailscale) for generation. 3-tier cache (memory → AsyncStorage → Supabase) for instant re-generation. Built a custom dut-recipe-generator Python service (560M BLOOM model, ~8s inference, 100% JSON reliability) as an alternative backend. Includes an AI adaptation loop, dietary preferences, skill level settings, and a community explore page with 44+ seeded recipes.
I like building things
During my time in university, I attended 2+ hackathons. People from around the country would come together and build incredible things in 2-3 days. It was eye-opening to see the endless possibilities brought to life by a group of motivated and passionate individuals.
Liquid AI × DPhi "AI in Space" Hackathon
Remote
Built SatChat: a vision LLM triage pipeline for satellite imagery. Fine-tuned LFM2.5-VL-450M with LoRA on Sentinel-2 agricultural stress data. 50,000:1 compression ratio over raw imagery downlink.
GT STAR-AI Makerspace Hackathon
Atlanta, GA — Parker H. Petit Institute for Bioengineering
Drug-target affinity prediction with GIN ensembles and uncertainty quantification. KIBA CI=0.882. Full paper, poster, and reproducibility package.
Get in Touch
Want to chat? Just shoot me a dm with a direct question on twitter and I'll respond whenever I can. I will ignore all soliciting.