Software EngineeringApplied AIData Systems
Computer Science + Business at UC Berkeley. I build AI applications with deterministic checks, resilient backend systems, and measurable impact.
Arrows move. Enter or Space reveals or chords a number. F flags. Right-click flags. Touch: tap reveals or chords; hold flags; two fingers pan or zoom. Activate the yellow smiley to reset.
First move is safe.
AI Research Intern
Built a FastAPI/PostgreSQL AI sales platform with human-reviewed outreach drafts and HubSpot integration, reducing manual prospect preparation by an estimated 80%. A live-call voice assistant reduced lookup time by 70%; worker leases, bounded retries, and a transactional outbox supported crash recovery and guarded against duplicate actions.
Software Engineer (Contract)
Built a Python/FastAPI and Next.js math tutor with exact-arithmetic checks and resilient PostgreSQL writes. Deterministic checks recorded zero false acceptances and false corrections across 85,942 generated cases evaluated against SymPy; maintained 2,600+ tests in GitHub Actions.
Generative AI Engineer (Contract)
Built a Python/LangChain evaluation harness running 35 financial-reasoning benchmarks per model checkpoint. Numerical and schema validators reduced manual scoring effort by an estimated 40% while retaining expert rubric review; failure analysis redirected data collection toward eight underperforming categories.
Director of Technology
Lead a six-person technology committee shipping React and Next.js features across Berkeley platforms serving more than 6,000 users.
Technology Consultant (Contract)
Modeled customer segments across 990 survey responses and translated the findings into three personalized-rewards product prototypes.
Software Engineering & Finance Intern
Built a Python ETL pipeline unifying three custodian export formats, cutting report preparation by approximately 40% across 75+ portfolios. Developed a LangChain pipeline that translated daily U.S. and Japanese news and mapped events to portfolio holdings.
Auditable AI finance workflows built for West Covina’s finance department.
Built a finance workflow platform for West Covina’s finance department to address untracked AI public-records risk. A fail-closed pipeline combines deterministic finance logic, source-linked LLM drafts, and audit logs. Eight finance workflows and natural-language Tyler/Munis export search cut ERP lookup time by 60%.
Co-founded an AWS-deployed college decision platform serving 7,600 unique users.
Co-founded College Explorer with a three-person team and deployed it on AWS, serving 7,600 unique users. Natural-language college search combines PostgreSQL full-text retrieval, structured filters, and reranking. A weighted ranker compares colleges across six fit dimensions, with data-coverage confidence and sensitivity analysis to make tradeoffs clear.
AI-assisted, event-driven backtesting platform with a natural-language strategy builder.
A from-scratch Python backtesting engine wired to FastAPI research APIs, a Next.js dashboard, and LangGraph research agents. Researchers draft strategies in natural language, run parameter sweeps and walk-forward validation, and analyze performance — all behind strict validation boundaries that never execute generated code.
Prediction-market intelligence dashboard correlating price, sentiment, and whale activity.
Vantage answers whether sentiment shifts and large-trader activity lead, lag, or coincide with prediction-market price moves. It ingests market data, detects anomalies, tracks whales, and runs on-demand sentiment analysis, surfacing explainable signals through a FastAPI backend and a React dashboard.
Deterministic tactical-arena simulator for reinforcement-learning research.
A replay-first, headless tactical combat environment for reinforcement learning and multi-agent behavior research. It runs 2v2 team-aware matches with configurable bot policies, parses natural-language commands into validated behavior profiles, and integrates PPO training through a Gymnasium wrapper.
Merged POSIX process-group cleanup into agent benchmarks, stopping timed-out process trees.
View on GitHub (Ponytail)Merged auto/MPS CLI overrides into the visual index builder, with pytest coverage for Apple Silicon.
View on GitHub (PixelRAG)I’m a Computer Science and Business student at UC Berkeley building reliable AI applications and data-intensive software. At Matrix Power, I built AI sales workflows and a live-call voice assistant. As a software engineer on contract with Google, I built a math tutor that pairs an LLM with deterministic checks. I’ve also built financial-reasoning evaluation infrastructure at AfterQuery and co-founded College Explorer, an AWS-deployed platform serving 7,600 unique users. I care about rigorous evaluation, explicit failure boundaries, and turning ambiguous problems into software people can trust.
Pursuing a B.A. in Computer Science and a B.S. in Business Administration at UC Berkeley and the Haas School of Business. GPA: 3.96.
View Résumé
