I've spent two years at the intersection of AI product development and user adoption.
As an Anthropic Campus Ambassador at UC Berkeley, I led a team that drove 10,000+ Claude Pro signups and built programming reaching 40,000+ students. I ran the frontline of how students discover and use Claude — organizing workshops, hackathons, and demos for hundreds of students. That work gives me pattern recognition most builders don't have: I see what excites new users, what confuses them, and where the gap between a model's capabilities and someone's actual workflow becomes a wall.
At SJF Ventures, I built AI tools for investors who had never touched a terminal — Salesforce automations, intelligent search engines for deal sourcing, and portfolio monitoring systems. I provided similar techincal assistance and advice to its porfolio companies in healthcare, education, and government, including developing some prototypes for new AI features. The constraint was the same: if the person using this can't understand it in 30 seconds, it doesn't matter how good the code is.
Now I build my own products. I prototype fast, test with real users, and iterate the same night. Every project below started with a pattern I noticed and a question I couldn't stop thinking about.
Projects
Things I've built.
AI Toolkit for Venture Capital
AI-powered deal sourcing and workflow automation for investors.
During my internship at SJF Ventures, no one on the team had an engineering background, and off-the-shelf AI tools didn't fit their workflows. I built a suite of internal tools: a Salesforce automation for logging portfolio updates, an intelligent search engine for deal sourcing, and research assistants that could surface relevant market data. I even developed prototypes and recommendations for new AI features for some of our portfolio companies. Every tool was designed for people who communicate in insights, not code.
ChatCHW - AI Health Assistant for Community Workers
Making healthcare information accessible for frontline community health workers.
Built an AI-powered health assistant for community health workers (CHWs) serving underserved populations. The system helps CHWs access medical information, answer patient questions, and navigate healthcare resources in real-time. Designed for users with limited technical literacy and unreliable internet access, prioritizing clarity and offline functionality over feature complexity.
Claude Showcase
A platform for students to share what they've built with Claude.
Ambassadors needed a central place to highlight student projects — demos were scattered across Slack threads and forgotten Google Docs. I built a showcase platform where students submit builds, browse what others have made, and get inspired to start their own. Shipped it to the ambassador community and received immediate adoption.
Postcards
A tool for sending personalized, storybook-style postcards to friends.
Maintaining close friendships after college is hard — everyone gets busy and "we should catch up" rarely turns into anything real. I wanted a way to reach out that felt intentional and personal, not just another DM. I built a digital postcard generator where each recipient gets a unique link to a storybook-style experience with pages for life updates, thoughtful questions, and shared memories. Shipped it to a close friend and it sparked a real conversation for the first time in months.
Writing
How I think about building.
November 2025
Yes, Even You Can Become TechnicalMy experience going from building no apps to building side projects every other week. A guide to when vibecoding works, when it breaks down, and the full stack I use to ship fast.
Read on SubstackSeptember 2025
I Vibecoded My Way Through My InternshipLessons from building AI tools at a venture capital firm where no one else had an engineering background. On choosing the right tech stack, prompt engineering as a skill, and building for the customer.
Read on SubstackApproach
How I work.
Observe first, build second
As a campus ambassador, I see what confuses and excites users every week. Every project starts with a real pattern — a pain point someone mentioned at a demo, a workflow that clearly isn't working, a question students keep asking. The best products come from paying attention before opening a code editor.
Ship fast, learn faster
Vibecoding collapses the feedback loop. I prototype in hours, put it in front of users, and iterate that same night. The code quality doesn't matter if the product is wrong — but once people are actually using something, that's when you invest in making it solid. Be wrong fast so you can be right faster.
Build for the person, not the prompt
My best work has been for non-technical users — investors, students, community members — who need AI to feel like a natural extension of their existing workflow. If someone has to think about the technology, you've already lost them. The interface should disappear.