Projects

From existing agents to reliable production AI.

Founding Software Engineer · Redwood City, California

Problem

Teams need to observe, evaluate, and improve agents built in different frameworks without rewriting them.

My contribution

Co-led the technical pivot from Maitai to Party, carrying the runtime, evals, and versioning forward into the new product instead of rebuilding. Built a structure-agnostic agent ingest and a schema-driven CLI import pipeline. Immutable revisions, pinning, and rollback keep edits separate from live agents.

Result

Agents from any framework land in a system with versioning and evals behind them. Dataset augmentation, embeddings, and model profiles lifted evaluation pass rates about 18%.

Legal document review at working speed.

Software Engineering Intern - Applied AI / Full Stack

Problem

Lawyers need to review large documents against legal guidelines, while retaining accurate citations beyond an LLM’s context window.

My contribution

Led multi-agent workflows with retrieval-augmented generation, and built the playbook review and summarize flows on top of them. Chunked inference across documents spanning millions of tokens so citations survived past any model context window. Introduced PostgreSQL views to reduce backend latency.

Result

Document review processed over 1,000 requests daily for hundreds of attorneys at midsize firms, and the backend refactor reduced latency by 70%.

Untangling identity services at scale.

Software Engineering Intern - Identity & Access Management

Problem

Critical identity endpoints were embedded in a monolithic application spanning more than 100,000 files.

My contribution

Decoupled Java endpoints into microservices, introduced rate limiting, and parallelized Cucumber acceptance tests with Spring Boot tooling.

Result

Reduced service latency by 60% and deployment time by 75%, while maintaining branch-coverage and mutation-test standards.