CS @ UC Irvine working across ML systems, AI research, and software engineering.

I like building where models meet systems: inference, evaluation, data pipelines, agents, and the infrastructure that makes AI useful beyond a demo. My work has ranged from production recommendation systems over hundreds of millions of records to LLM confidence research, scientific ML, robotics, and low-level GPU programming.

Portfolio · LinkedIn · Google Scholar · Resume

• 🏜️ Sandia National Laboratories — building full-stack applications and backend services that make AI research accessible through user-facing tools
• 🤖 Underwater Robotics @ UCI — developing perception software for autonomous underwater robotics
• 🔬 UCI Digital Learning Lab — researching LLM confidence and evaluation across multi-agent reasoning and multi-turn Text-to-SQL
• 🌡️ Calit2 @ UCI — developing physics-informed ML to replace expensive thermal simulations
• ⛓️ Blockchain @ UCI — building smart-contract security tooling with Foundry, Slither, and adversarial test contracts
• ♻️ Lazarus — building open-source recovery infrastructure for abandoned cloud-dependent devices by analyzing protocols and recreating the minimum services needed for local control
• ⚡ ContextCUDA — exploring custom C++/CUDA kernels for long-context LLM inference, including attention, RoPE, KV-cache management, and context efficiency

• Sandia National Laboratories — Software Engineer Intern
  Building full-stack applications and backend services that turn AI research into usable software.

• Capital One — Software Engineer Intern
  Built recommendation infrastructure over ~800M customer-response outcomes, including a 28× faster Databricks sampling pipeline and an SDK migration that reduced serving latency by 25% for systems reaching 60M+ users/month.

• Underwater Robotics Project @ UCI — Software Engineer, Perception
  Developing perception software for UCI's autonomous underwater robotics team.

• Blockchain @ UCI — Technical Developer
  Building AI-powered smart-contract security tooling; previously developed autonomous agent infrastructure for the decentralized AI trading platform Agonus.

• CareTech @ UCI — Software Developer
  Built a real-time food recognition and nutrition system with PyTorch and FastAPI, processing 28K+ images across 270+ food categories and training models to 81% test accuracy.

• Commit the Change @ UCI — Full-Stack Developer
  Built healthcare scheduling and provider infrastructure supporting 8,000+ patients/year and saving staff 12+ hours/week.

• SENS Psychology — Software Development Intern
  Built clinical intake and operations software used across 1,200+ monthly patient interactions, cutting intake time 50% and manual emails 75%.

• PROP.INTEL · IrvineHacks 2026
  AI property-risk platform combining property records, FEMA disaster data, and machine learning to generate real-time risk analysis for U.S. properties; modeled exposure across 3,100+ counties and 8 hazard types

• Agonus · Blockchain @ UCI
  Decentralized AI trading platform with autonomous LangChain ReAct agents, on-chain tournaments, and Celery/Redis infrastructure for reliably orchestrating 10+ concurrent agents

• EthicsChat · Tata Consultancy Services
  Ethics-gated financial chatbot combining a DeBERTa moderation layer with an end-to-end RAG pipeline for document-grounded responses

• GitHub Onboarding Agent · Personal Project
  Developer onboarding system that ingests unfamiliar repositories and provides code-aware exploration, semantic search, and grounded RAG Q&A

UCI Digital Learning Lab

Researching confidence, reliability, and evaluation in LLM systems, with a focus on token-level decoding confidence, multi-agent reasoning, LLM-as-judge evaluation, and multi-turn Text-to-SQL.

• The First Tokens Matter: Early Confidence Signals for Evaluating LLM Reasoning — ICLR 2026 LLM Reasoning Workshop
• The Confident Liar: Diagnosing Multi-Agent Debate with Log-Probabilities and LLM-as-Judge — ACL 2026 Student Research Workshop
• Early-Token Confidence Predicts Reasoning Quality in Multi-Agent LLM Debate — GEM 2026

Calit2 @ UCI

Developing physics-informed neural models as fast surrogates for computationally expensive thermal simulations. Built a physics-informed GRU constrained by heat-transfer equations, achieving R² = 0.99, 1.54°C MAE, and roughly 12,000× faster inference than FEA simulation.

Languages

ML / AI

Data / Scientific Computing

Backend

Systems / Cloud

Frontend

Learning about marine life and ocean creatures 🐠 · playing basketball 🏀 · finding good food 🍜