San Diego, CA · Building Audible Sight
Engineer & CTO. I build AI products people can actually use — and ship them all the way to production.
Currently CTO of Audible Sight, where we make video accessible to blind and low-vision audiences. Before that: ML engineering, AI consulting, and a Math–CS degree from UC San Diego.
San Diego by way of Singapore. Math–CS at UCSD. Cooks, solves puzzles, ships things.
Read more → WorkAI audio description, enterprise SSO and compliance, LLM evals, and weekend builds.
Browse projects → PlayThe "fun facts" section is a playable Connections board. Four groups, four mistakes.
Solve it →I like the whole stack of a problem — not just the model or the code, but the infrastructure underneath it, the compliance paperwork around it, and the person on the other end using it. As a founding CTO, I've had to own all of it: designing AI pipelines, deploying them on cloud infrastructure, getting them through enterprise security review, and refining the product until it earns its users.
My background is Mathematics–Computer Science at UC San Diego (3.9 GPA, Provost's Honors), with stops in ML engineering and AI consulting along the way. The thread through everything: technology should widen who gets to participate — whether that's video for people who can't see it, or coding classes for kids who'd never been offered one.
I grew up between worlds — family in Singapore and India, education and career in California — and I think it shows in how I work: comfortable switching contexts, allergic to assuming everyone shares my defaults, and happiest when something I built travels further than I expected.
Off the clock I cook (genuinely, not aspirationally), build side projects, and protect a Wordle streak with unreasonable seriousness.
A demo isn't a product. I care about the unglamorous last mile — auth, monitoring, compliance, docs — because that's where software becomes real.
Accessibility isn't a feature checklist; it's a design constraint that makes everything better. The best products work for the people most tools forget.
From RA life to CSforEACH to a Python textbook used by 1,000+ students — explaining things clearly is half of understanding them.
Startups, internships, and a university chapter that did a lot of heavy lifting.
Leading engineering for an AI audio-description platform serving blind and low-vision audiences.
LLM evaluation and monitoring for a production customer-experience AI platform.
Client-facing AI engineering, from technical discovery to production deployment.
3.9 GPA, Provost's Honors.
Production systems, trust-building infrastructure, and the weekend builds that keep it fun.
The core product: intelligent frame selection, GPT-4o scene description with a character registry that keeps names consistent across a film, and natural TTS narration. Frames in, voice out. In production for paying B2B customers.
The editor where humans refine the machine's words — synced video player, auto-scroll, and a transcript format that merges dialog and description into one readable track.
Took a startup to enterprise-ready: SAML SSO via Auth0 and Microsoft Entra for government customers, a full CAIQ self-assessment for CSA STAR Level 1, and the security policy suite behind it.
Production-grade observability at 1M+ traces a month: LLM-as-a-judge scoring, safety signals, and dashboards that catch drift before users do — lifting response accuracy by 18%.
An Outlook plugin that classifies and routes email for a 100+ user client — cutting processing time 45%, with a validation layer holding 85%+ intent accuracy in production.
A rotating cast: a bot that solves NYT Connections, an AI file-organization agent, and whatever idea is currently winning the fight for my Saturday mornings.
I solve Connections every morning, so the "fun facts about me" section is one. Sixteen tiles, four categories, four mistakes. One of these is sneaky on purpose.
Word games are my daily warm-up — I even built a bot that solves Connections. Puzzles and engineering scratch the same itch: constraints, patterns, and the small joy of a clean solve.
The fastest way to reach me is email — I read everything, and I reply to anything with substance. If you're working on accessibility, AI products, or just have a hard problem that needs shipping, I'd love to hear about it.