$ whoami
_
AI Researcher & Engineer
Interested in Healthtech Agentic AI AI Infrastructure Simulation Software
01. About
I build intelligent systems that push the boundaries of what's possible.
Currently pursuing EECS at UC Berkeley while researching AI safety with Prof. Dawn Song and Prof. Claire Tomlin. Previously shipped production AI at Apple and JustPaid.AI (YC W23).
My work spans multimodal ML, agentic systems, and safety verification: from building CLIP-powered search tools to developing Hamilton-Jacobi algorithms for autonomous systems.
02. Experience
Software Engineering Intern
AppleBuilt AI-driven text-to-image search for Apple Ads using multimodal CLIP. Greenfield initiative, reduced relevancy search from days to seconds.
AI Engineering Intern
JustPaid.AI YC W23Designed and shipped agentic LangGraph workflow for one-click traceability. Automated validation slashed manual review time by minutes per document.
AI Research Intern
NASA / UT AustinDeveloped transformer and CNN models achieving 65% accuracy improvement. First-authored paper on arXiv. Presented at AGU Fall Meeting.
Research Intern
Stanford AIMI CenterImproved endotracheal tube prediction accuracy by 42% with novel CV models. Selected as top 2% of 1,100 applicants.
Researcher
MIT BeaverworksOne of 25 selected from 4,000 for Autonomous Cognitive Assistant program. Built music identification, image search, and face recognition models.
03. Research
Agentic Benchmarks
Prof. Dawn Song
Building novel benchmarks for end-to-end website screenshot to code generation. Evaluating frontier models on complex agentic tasks.
Active ResearchAI Safety Verification
Prof. Claire Tomlin
Developing data-driven Hamilton-Jacobi safety verification algorithms for controlling systems with unknown dynamics.
Active ResearchMosquito Larvae Detection
NASA / UT Austin
Transformer and CNN-based image classification for disease vector surveillance. 65% improvement over baseline models.
Read on arXiv →04. Projects
Eloquence
Real-time ASL translation web app. Trained and deployed MediaPipe hand-tracking ML model on custom dataset of 400+ images.
05. Get in Touch
Building something interesting? Let's talk.
$ echo $EMAIL
aswinsurya@berkeley.edu