University of California, Berkeley
GPA 3.8/4.0
- Optimization Models (EECS 127)
- Probability & Discrete Math (CS 70)
- Deep Learning (CS 182)
- In progress: Probability & Random Processes (EE 126), Linear Algebra (Math 110), Algorithms (CS 170), Machine Learning (CS 189)
AI/ML & Quantitative Research Engineer
- Built a reinforcement-learning system (DQN: reward shaping, experience replay, target networks) and the full experiment + data pipeline (SvelteKit/Firebase) over 85 participants and 1,000+ rounds.
- Clustered behavior into 5 statistically separable decision policies (8× outcome spread, F=276) and ran offline policy evaluation.
- Fixed-effects causal inference showed payout salience raised suboptimal choices by 71 percentage points with no measurable learning across 15 rounds.
- First author, in preparation (targeting ACM CHI 2027); presented with Prof. Sinchaisri at Stanford and CMU.
Python
PyTorch
JavaScript
TypeScript
FirebaseData Engineering Intern
- Built and hardened large-scale equities market-data pipelines across multi-terabyte tick and daily-bar feeds with billions of rows.
- Optimized Spark/Pandas processing over partitioned Parquet for 20% faster runtime and automated ETL with Airflow.
- Tuned PostgreSQL performance and supported Redis caching and Dockerized services for production data workflows.
Python
SQL
Pandas
NumPyPostgreSQL
DockerMachine Learning Researcher
- Built a deep generative + reinforcement-learning stack in PyTorch with diffusion sampling and policy/value methods over a multi-objective reward.
- First-authored a peer-reviewed paper on AI-driven molecular design (DOI: 10.36838/v7i4.29).
PythonPyTorchNumPyPandasComputational Research — Ranked 1st of 45
- Built Monte Carlo, stochastic-process, and time-series models of nonlinear biochemical systems.
- Implemented PCA, k-means, and random-forest pipelines for computational research workflows.
Python
C++NumPyPandas
scikit-learn
Research Platform
BundleGame
SvelteKit and Firebase research app for studying delivery-order bundling decisions across a 50-round task, with participant telemetry, Qualtrics linkage, analytics exports, and offline-RL artifacts.
AI Safety Product
Manyr Agent Firewall
Next.js product site and interactive policy playground for an agent-governance layer that evaluates tool calls, applies allow/deny/approval rules, and records audit trails.
Full-Stack Pipeline
CivicGrid
Voice-to-structured-data civic triage platform using Deepgram STT/TTS, LLM classification, Firebase, and a role-based dashboard for residents, contractors, and officials.
Quant Research
Cross-Sectional Equity Signal Research
Point-in-time S&P 500 study that caught and killed a survivorship artifact: a +8%/yr factor alpha (t ≈ 3.6) vanished under point-in-time membership; honest null with deflated Sharpe ≈ 0.35, independent re-derivation, and 50 unit tests.
Portfolio Website
Personal Website
Next.js portfolio site with interactive 3D visuals, smooth scroll animation, project cards, contact flow, SEO metadata, and responsive presentation for AI/ML and quant research work.
