Senior Staff Engineer at Meta · Voice AI & Audio Infrastructure · Founder
I'm a systems engineer building at the intersection of real-time audio, on-device ML, and AI infrastructure. At Meta, I lead the Voice AI Platform -- an 8-stage streaming audio pipeline that powers always-on ambient intelligence on Ray-Ban Meta smart glasses, from dual-microphone capture at 48 kHz through privacy-preserving transcription in trusted execution environments to multi-modal retrieval over weeks of conversational memory.
Outside Meta, I build products. I'm the founder of three startups spanning AI-native documents, private social sharing, and enterprise data infrastructure. I care about shipping things that work, building systems that scale, and the craft of turning research into products people use every day.
AI-native document platform for agents and humans
HTML Docs is a document platform built for the agent era. Any AI agent can publish a polished, hosted web document with a single API call -- no auth required, instant URL, full editing and collaboration from the moment it lands. The thesis: as AI agents become primary content creators, they need a native output layer that goes beyond markdown dumps and terminal logs. HTML is the new markdown.
The platform serves both sides: agents get a clean POST-HTML-get-URL API, MCP server integration, and a forward proxy for building rich documents. Humans get a full editing environment with real-time collaboration (Liveblocks), AI-powered Docsmith chat for in-document editing, native PDF viewing with annotation, version history, and Google Docs export. The product supports Claude Code, OpenClaw, Codex, and any agent that can make an HTTP call.
Key infrastructure: Next.js on Vercel (SFO region), Supabase (Postgres + Realtime + Storage), OpenAI for Docsmith, server-side Mermaid rendering via Kroki, client-side PDF engine via pdfjs with custom column-aware text selection.
The private rolodex for sharing memories with friends
Amika is a private, intimate social app designed for the people who actually matter -- your close friends. Think of it as a digital rolodex meets shared journal: a place to collect and revisit memories, moments, and inside jokes with your inner circle, away from the noise of public social media.
The app is built around the idea that the most meaningful social interactions happen in small groups. Instead of broadcasting to hundreds of followers, Amika creates private spaces where friends can post memories, react to shared moments, and build a living archive of their friendship. It's the anti-feed -- no algorithms, no strangers, no performative posting. Just the people you care about and the moments you want to remember.
Customized ETL for enterprise marketing data
Skybright builds custom data pipelines for enterprise marketing teams that have outgrown off-the-shelf ETL tools. The core problem: marketing data is fragmented across dozens of platforms (ad networks, CRMs, attribution tools, analytics suites), each with its own schema, API quirks, and rate limits. Generic connectors break constantly; marketing teams end up with stale dashboards and unreliable attribution.
Skybright solves this with bespoke ETL pipelines tailored to each customer's specific stack and data model. The platform handles extraction from all major marketing platforms, normalizes schemas across sources, resolves identity and attribution across touchpoints, and delivers clean, unified datasets into the customer's warehouse. The result: marketing teams get a single source of truth for spend, performance, and attribution -- updated reliably, without engineering babysitting.