Building Meta’s AI Coding Agent from the ground up. I architected & executed the first large-scale LLM-based code refactor ever performed at Meta. Built several large-scale software platforms from scratch. Got 400,000+ user signups. Improved iMovie and Final Cut Pro as a software engineer at Apple.
Menlo Park, CA
Building Meta's AI Coding Agent, Devmate, from the ground up. Zuck said 50% of new code at Meta could be written by Devmate in the next couple years.
I personally:
Hooked up native tool use / function calling across all models
Managed all LLM inference spend and rate limiting
Built orchestration layer for our terminal agent
Last year, I architected and implemented a system to refactor a multi-million line-of-code codebase using an LLM. It was the first large-scale AI code refactor ever performed at Meta, and received public praise from a senior eng director.
2022 — 2024
Sunnyvale, CA
Financial planning platform with sophisticated financial projections and game-like drag-and-drop UI
Features: Organize all of your finances. Create and manage your financial objects, which include generic object types like Asset, Loan, Expense, Income and Folder, along with 40+ specific object types like Child, House, Student Loan, Mortgage, 401k, Social Security, etc. Link your bank account to keep your financial picture up to date and organize your transactions. Plan your finances into the future. View and project your Cashflow statement and Worth statement (Balance Sheet) and Taxes (Federal Form 1040 + State Taxes) into the future.
Implemented a version control system and a "sync" algorithm to sync changes from the financial picture (current financial state) into plans
Optimized stats recalculation algorithm by implementing a cache mechanism so the app only recomputes stats that are affected by new changes - cut down recomputation time 40X - for context, the app computes "worth", "change in worth", and "cashflow" for every object in your financial plan, month-by-month, 80 years into the future
Built app with Typescript, React Native, Reanimated, Node.js, MongoDB, and AWS - deployed backend infrastructure and codebuild pipelines using AWS Cloud Development Kit (infrastructure as code) utilizing 14 popular AWS services
Collaborated with the Reanimated (React Native animation library) open source development team to fix bugs in Reanimated - they gave Piggy priority bug support because the app was one of the heaviest usages of their library
San Francisco, California, United States
User Research & Survey Platform - Series B Startup - #1 Most upvoted product on Product Hunt in 2022
Architected and implemented the "Cohorts" feature which enabled delivering targeted surveys to groups of users
Wrote mission-critical backend code that processed thousands of requests per second and selectively delivered surveys based on boolean logic rules (using Node.js, Postgres and Clickhouse)
Generated SQL queries for logic rules that used multilevel boolean operators with dozens of clauses and aggregations
Shipped over 80 PRs in a 2 month period, working closely with a Staff backend engineer
2019 — 2021
Web and Mobile App Platform to get users bank fee refunds, rebates, coupons and rewards from online stores
Lead all development: backend (Node.js), database (MongoDB), AWS infrastructure, security across all surfaces, desktop and mobile-friendly web apps (AngularJS), Android and iOS apps - 4.7 stars with 2,500 ratings (Swift, Java)
Scaled app to 400,000+ user signups, $90k monthly revenue, 400+ million database records, 41 database tables, 63k lines of custom JavaScript code, 148 NPM packages, 91 transactional emails, 23 third-party APIs, managed 2 engineers
Automated backend systems to process "claims": bank fee scanning, email receipt scanning, submitting correspondences with banks and merchants via physical mail (Lob API) and email (Gmail, Yahoo! (IMAP), and Microsoft APIs) to get refunds, rebates and rewards for users
2018 — 2019
Cupertino, California, United States
iMovie & Final Cut Pro (Video Apps)
Collaborated with the Data Model team to improve iMovie and FCPX's project import, export and sharing capabilities
Built features and debugged in the extremely complex, 20M+ line-of-code FCP/iMovie codebase - made changes at the lowest levels of the application, thousands of nested function calls deep
Made frequent changes to address concurrency issues, modify the data model and manage media files on the filesystem
Increased the efficiency of debugging by 3x for a team of 6 engineers working on the next major Final Cut Pro X feature by implementing a smart debug alert that surfaced the low-level root causes of most bugs
Partnered and pair programmed with Principal Engineers on mission critical projects
Education
Santa Clara University
Computer Science and Engineering
Phillips Exeter Academy
Fullstack Academy