I am a full-stack developer and AI engineer finishing my MS in Computer Science at Rutgers, shipping production platforms and AI tools people rely on.
Open to SWE & AI/ML rolesBased in New Jersey
Motion access is blocked for this site. Turn it on in your phone’s settings for this browser — look for Motion & Orientation Access — then reload.
ScrollSelected work
Helps blind users find lost items. Just say “Hey BlindSpot” to give commands, and a multi-step, prompt-based agent starts streaming live camera frames through Gemini 2.5 Flash, speaking the directions back in real time — “your keys are at 2 o’clock”.
AwardBest Community Impact Award — Rutgers Raptor Rise, out of 50 projects
Built withPython · FastAPI · Gemini 2.5 Flash · ElevenLabs TTS · Computer vision · Prompt engineering
Answers domestic workers’ questions about their legal rights in plain language, scoped to their own state or city. Government policy PDFs are semantically chunked, embedded and searched in Pinecone, so every answer comes from a verified document.
Award2nd place, HackRUC 2025
Built withPython · FastAPI · Next.js · TypeScript · Gemini 2.0 Flash · RAG · Pinecone · MongoDB
A cross-platform fitness and nutrition app, designed and built end to end. An AI coach answers with your day’s training and meals in context, a meal photo returns its macros, and a recovery engine scores every muscle group from published training-load research.
Under the hoodEvidence-based recovery engine — deterministic per-muscle fatigue scoring, with the papers cited in-app
Built withReact Native · Expo · TypeScript · Python · FastAPI · MongoDB · Firebase · Gemini Vision · Claude Haiku
Maps supplier concentration risk across Aramark’s 2025 US procurement spend — 43M rows in BigQuery. No state has a competitive distributor market, weather measurably shifts what gets bought, and k-means finds six buying archetypes.
FindingOne distributor controls 54% of US spend — antitrust risk quantified with the Herfindahl-Hirschman Index
Built withPython · pandas · BigQuery · Google Cloud Storage · scikit-learn · k-means · NOAA data
Treats the whole AWS EC2 price list as a dataset — 318,604 prices, 1,337 instance types, 104 regions — to ask how AWS really prices compute. Pareto testing finds the 6.2% on the efficiency frontier, and PCA projects the catalogue into two dimensions.
FindingThe saving is Graviton, not a newer generation — 15–20% cheaper than Intel at matched generation and size
Built withPython · pandas · NumPy · scikit-learn · PCA · Plotly · Jupyter
About
I pick the problem first and learn whatever it takes to build it, and the projects I care about most started the same way. A few months ago I went to an assistive-technology workshop and got talking to the speaker afterwards. She told me that when a blind person loses something at home, their only real option is to video-call a stranger and ask them to look through the camera. I could not stop thinking about it. BlindSpot came out of that conversation — the award came later, and it was never the point.
That is the pattern: I build things nobody asked for, and the findings I am proudest of came from questions nobody assigned me. Right now I am finishing my MS in Computer Science at Rutgers and looking for somewhere to keep doing it.
Away from the screen I train, I read, and I spend a probably unreasonable amount of attention on habits — the small things you repeat until they compound into something. It is the same instinct as the software: nothing good arrives finished, you just keep making it slightly better than it was.
Experience
Built and deployed two production web platforms (real-estate lead-gen and B2B industrial e-commerce). Integrated Stripe Checkout & webhooks, Google Maps, Firebase, and email APIs for $10K+/mo billing and live business workflows.
Built a Siamese Autoencoder intrusion-detection system on NSL-KDD (148K+ samples) in PyTorch with contrastive loss and curriculum learning, lifting accuracy from 54.7% to 82.1%. Authored two research papers.
Guided 25+ students on input validation, sanitization, and authentication in Python and TypeScript; ran Wireshark and Burp Suite lab sessions.
Worked with 1M+ rows of sales data in Python, applying regression and classification models; cut preprocessing time by 30% and delivered EDA reports for the marketing team.
Contact
Open to software engineering, full-stack and AI/ML roles.
