I love pushing AI
to the frontier.
I'm 20, building at the edge of AI: agentic systems, LLM evals, and applied research. Before this, I shipped agentic NLP at ASAPP, built a neural-net tool 500+ students learned on, and co-authored LLM research headed to ASN 2026.
ColdBrew
🥇 1st place, HF0 Voice & Video AI Hackathon. 1 of 30 builders, picked from 1,000+ applicants. I almost skipped it because I couldn't afford the trip. I went anyway.
Thursday 9PM I got in. Friday I flew across the country. Saturday my team won.
We built real-time vision for warehouse cameras: it flags safety violations, equipment failures, and shipment anomalies live, on NomadicML's vision API. Twelve hours, with people I met that morning. We demoed it to VCs and founders at the Archbishop's Mansion.
If you're a student who thinks you can't afford to be in these rooms: book the flight, figure it out on the way. The worst thing you can do is not show up.
Extractify
Agentic NLP I built during my fellowship at ASAPP. It turns messy customer conversations into clean, structured data.
- 10K+ multi-turn chats, 95-100% accuracy on key fields
- Hybrid regex + LLM pipeline on LangGraph and Gemini 2.5 Flash
- Pulls entities, intent, sentiment, and summaries
- Benchmarked it myself: PCA, KMeans, KNN, F1 > 0.85
Presented the results to ASAPP's ML leadership.
NeuroBlock
The thing I'm proudest of. Drag-and-drop neural networks in the browser: build, train, and export classification, regression, and CNN models with no code.
500+ high schoolers have used it to actually get how neural nets work. Training runs stream live over WebSockets, so you watch the model learn.
Most ML lessons open with math that scares people off. This one opens with a block you drag.
// where I've been
- Founders, Inc.Mar - Jun 2026Founder, Canopy · San Francisco
- Selected for Canopy, the Founders, Inc. founder program.
- Hubert Thieblot brought me on to build on-site in SF.
Canopy0→1SF - AssemblrDec 2025 - 2026Co-Founder
- Workflow intelligence for engineering teams. Agents built from how teams actually work.
- Top 10% at YC. Top 10 at Speedrun.
agentsproductSpeedrun - ASAPPSep - Dec 2025AI Engineering Fellow
- Built Extractify: an agentic NLP system on LangGraph + Gemini 2.5 that pulls clean structured data from messy customer chats. 10K+ conversations at 95-100% accuracy.
- Hybrid regex + LLM pipeline for entities, intent, sentiment, summaries.
- Benchmarked it with PCA, KMeans, KNN (F1 > 0.85). Presented to ML leadership.
LangGraphGemini 2.5PythonML - MIT, Break Through Tech AIApr 2025 - May 2026Fellow
- Competitive, fully-funded AI/ML fellowship. Real industry projects and mentorship.
fellowshipAI/ML - Susquehanna (SIG)Dec 2025Discovery Day Fellow
- Quant trading, probability, and engineering problems in financial markets.
quantfellowship - Advanced Learning Tech LabFeb 2025 - Jan 2026Undergraduate Intern
- Built a multi-agent pipeline over Hugging Face APIs for K-12 wearable learning data.
- Automated CSV-to-prompt generation. Cut parsing time 60%.
multi-agentHugging Faceresearch - ML for Education LabSep 2024 - Apr 2025Research Assistant
- Tested GPT-4.5, Gemini 2.5, Claude 3.7, and DeepSeek-R1 against clinical dietary guidelines for accuracy and bias.
- Co-authored a paper accepted to ASN 2026.
LLM evalresearchpublished - PaktolusMay - Jul 2025Software Engineer Intern · Florida
- Built a full-stack user management system: Django REST + FastAPI + Next.js 15, with role-based access.
- JWT auth, account lockout, file uploads. Test coverage up 80%+.
Django RESTFastAPINext.jsTypeScript - telSpielDec 2024 - Jan 2025Software Developer · Noida, India
- Built a React dashboard. Cut navigation time 40% for ~3,000 daily users.
- Automated DLT tracking, cut compliance errors 35%. On a system handling 150M daily interactions.
Reactscaledashboards - CICSoftFeb 2025 - Feb 2026President
- Ran the biggest tech org on campus.
- Spent most of it helping people get into rooms they didn't think they belonged in.
leadershipcommunity
// what I'm into right now
- LLM evals & reliabilitywhat models actually get right, measured
- agentic systemsmulti-agent pipelines that do real work
- frontier models & reasoningthe edge, before it becomes a trend
- on-device / edge MLintelligence where the compute isn't
- applied AI researchpapers that turn into products
// what I'm about
- I don't chase trends. I'd rather be early to the thing that becomes one.
- I love the engineering. I love the customer more. ❤
- I put my whole heart into everything I build.
- I'd rather take a real swing at a huge problem than ship a 24-hour gimmick.
ps. i'll be at YC Startup School. hmu :)
// say hi. short, specific notes get the fastest reply.