• Led a 7-person engineering team to design and build an AI Coding Assistant that enables users to generate and iteratively refine full-stack web applications through natural language and LLM-driven code generation.

• Engineered Webflow Cloud’s full-stack hosting platform, deploying applications on Cloudflare Workers with scalable, reliable infrastructure for modern web app development.

• Improved reliability and safety of LLM-powered features by building evaluation frameworks, LLM observability, and tracing systems that raised output quality and user trust in AI Features.

• Independently architected, developed, and launched Resonate, a fullstack GPT-4 powered user research assistant that instantly transcribes interviews and creates and shares interview highlights/snippets.

• Architected and developed the system using React frontend, Node.js, GraphQL API, and a Python Flask service with fully automated CI/CD and observability and 99.99% uptime.

• Developed features to enable users to tag user interviews during Zoom calls in real-time and automatically identify patterns across multiple user interviews by continuously aggregating common tags and themes with GPT-4; reducing required manual working time for end users from weeks to minutes.

• Redesigned the frontend architecture of a React application using an internal frontend architecture to enable the development and deployment of 2 new web dashboards/portals for new customer segments while maximizing reusability.

• Streamlined communication between implementation teams and legal domain experts and reduced time to launch for new customers by 20% by creating a Domain Specific Language for coding complex debt financing contracts.

• Doubled multimillion-dollar MRR by leading Trello’s 7-person funnel expansion team through monetization projects that introduced a limit on the number of boards for free teams and prevented teams from adding free guests on >1 board.

• Built patented A/B testing deployment and analysis infrastructure consisting of libraries for bucketing and tracking users across all Trello clients, an event measurement system, and a Python analysis service that has been used to deploy and analyze hundreds of A/B tests across Trello’s fullstack; becoming the de-facto method for releasing new features.

• Reduce time to first render by 80% in BitBucket by overhauling the performance of large pull requests with a 4-person team