Applied AI · Agent Systems · LLM Reliability
I build production-ready AI systems that combine LLM reasoning, structured tool-calling, and human-in-the-loop workflows. Based in Daly City, CA.
AI email agent that triages, summarizes, and drafts replies using LangGraph.
- LangGraph agents with structured tool-calling (Pydantic schemas) and deterministic execution
- Tiered RAG retrieval: Redis (hot storage for active threads) + SQLite (cold storage for archived context)
- Stateful workflows with checkpointing, retry logic, and execution logging
- Gmail + Slack Bolt integrations with interactive modals
- Human-in-the-loop approval system
- Productionized as a Flask service with modular routes and environment-scoped secrets
- Treat LLMs as unreliable components → enforce structure, validation, and fallback paths
- Prefer deterministic execution layers (schemas, tool constraints) over prompt-only control
- Design for observability: logging, replayability, execution tracing, and failure inspection
- Optimize for iteration speed via eval loops and real-user feedback
ML Engineer (Applied AI) · Inbox0 (May 2025 – Present)
- Architected production LangGraph agents with schema-validated tool-calling and HITL approval
- Built tiered RAG system reducing context-lookup latency while keeping full email history queryable
- Implemented evaluation infrastructure: execution logging, retry logic, and state checkpointing
Open-source Developer · sktime (Mar – Oct 2024)
- Decoupled GridSearch from scikit-learn, reducing user-reported issues by 20%
- Shipped
MultiplexerRegressorAutoML component — cuts model-selection time by 30%
AI Training Engineer (RLHF) · Self-Employed (Feb – Nov 2024)
- OpenAI (ChatGPT) — evaluated reasoning, tool use, and prompt adherence across multi-turn conversations; wrote Python tests to verify generated code matched user intent
- Google (Gemini) — rewrote chain-of-thought traces to compress agentic turn counts; validated tool-calling schemas for Maps, Search, Calendar, and Purchasing
- Meta (Llama) — scored output quality, penalized hallucinations, and re-evaluated downstream responses after failure-point regeneration
Data Scientist · Acorn Analytics (Aug 2020 – Mar 2021)
- Built Python/MySQL AWS ETL pipelines; ran 10K Monte Carlo simulations for headcount planning
- Collaborated on a logistic regression fraud-detection model → $250K cost savings
Staff Consultant · Celerity Consulting Group (Mar 2016 – May 2018)
- Led team of 5 analysts; built Tableau dashboards via MySQL on SharePoint ETL pipelines
Stanford University (2022 – 2024) CS229 (Machine Learning) · CS224N (NLP) · CS231N (Computer Vision) · CS336 (LLMs) · CS236 (GANs)
UC San Diego (B.A. Cognitive Science, 2015)
- Building more reliable agent workflows — retry strategies and state recovery
- Exploring lightweight models to reduce cost in agent pipelines
- Improving evaluation frameworks for agent behavior
Python LangGraph Pydantic AI PyTorch OpenAI APIs RAG RLHF Flask AWS GCP SQL Redis Linux
ML / AI engineering roles focused on agent systems, LLM applications, and evaluation.