I build at the intersection of machine learning and systems — training models, deploying them to real edge hardware, and caring about what happens all the way down to the compiler.
🔭 Currently working on — a personal project that tracks wildfires and grinding DSA for new-grad SWE interviews
👀 Open to — new-grad Software Engineer & ML-systems roles
🛠️ I build — applied-ML systems that run on constrained hardware, not just in a notebook
🌱 Currently learning — how to use agentic workflows
💬 Ask me about — running YOLO inference across two Jetsons, ZeroMQ messaging, or building a compiler from scratch
⚡ Fun fact — I admin a modded Minecraft server for a community of friends
A full-stack price-comparison app: a Playwright scraper pulls live per-store grocery prices from Instacart into Supabase (Postgres), and a native SwiftUI iOS app turns any recipe into the cheapest — or fewest-stores — shopping trip. Recipe ingredients are fuzzy-matched to real products offline (rapidfuzz + Claude API reranking for near-matches), so the app never matches on the hot path. I led the four-person team and owned the scraping pipeline (persistent browser sessions, bot-detection handling, idempotent upserts) and the Supabase data architecture.
Real-time produce-quality inspection running distributed YOLOv11n inference across two NVIDIA Jetson Orin Nano edge devices. Trained on a merged 26-class, ~10,750-image dataset; ZeroMQ PAIR sockets over direct Ethernet, Flask web streaming, and Google Cloud Storage integration. I owned the training pipeline, dataset aggregation, and model versioning — and debugged the real-world OOM, socket desync, and camera/inference contention along the way.
A multi-phase compiler written in Rust — lexing and parsing through IR code generation, with control flow (while, if/else, break/continue) and a working symbol table. A from-scratch look at the machinery languages are built on.
NLP components built without leaning on high-level libraries — to understand the math, not just call an API. Word2Vec (SGNS) in PyTorch, a logistic-regression sentiment classifier, n-gram language models, and a statistical spellchecker (Damerau-Levenshtein + bigram LM).
