-
Sift's autonomous goal agent — a post-synthesis AI agent (built on Mastra) that reads every inbound customer conversation, picks the right org-defined support goal (or abstains), and executes end-to-end: drafting replies, tagging, assigning, escalating, closing cases. 4.9K+ autonomous runs and 3.5K+ executed goal decisions across 19 social platforms — X, Instagram, WhatsApp, Reddit, Discord, TikTok, and more.
TypeScriptMastratRPCHatchetPostgreSQL -
Sift's real-time synthesis pipeline — enriches every ingested message with sentiment, relevance, actionability, topics, and themes: the structured signal the agent reasons over. Scaled to 2.9M+ AI-enriched records out of 31M+ ingested messages (~2K enrichments/day), with cost-control batching that keeps LLM spend predictable at scale.
BAMLGeminiInngestTimescaleDBKysely -
Graphite — Google Summer of Code 2026 contributor to the Rust-based open-source 2D vector graphics editor. 14 merged PRs (7 more in review): correctness fixes across the Path/Shape/Line/Freehand tools, undo/history bugs, rasterize-node document leakage, off-by-one PNG exports, Firefox color rendering, F2-rename and Ctrl+J path joining.
Rust -
Prisma — merged fix to the Prisma Client JS generator: closed an unterminated comment block in the generated file preamble that broke the
biome-ignore-alldirective, which was flooding create-t3-app users with hundreds of Biome lint errors on generated files.TypeScript
I'm a founding engineer at Sift with 2+ years building production AI/LLM systems that run at scale — not just demos.
I designed Sift's agentic workflow system and the synthesis pipeline that feeds it, shipped services across a multi-target deployment stack (Next.js on Vercel, Dockerized Mastra AI/MCP services on AWS Fargate via SST + Flightcontrol, a Cloudflare Workers gateway), and operate it all with Datadog APM, Infisical secrets, and Inngest + Hatchet for durable orchestration. Before that I was a software engineer at Amazon on the Aurora Global Databases team, and an analyst at Bain & Company (BCN Labs). I've also contributed merged fixes to Prisma and Clerk.
I live at the intersection of LLM orchestration, backend services, and data-intensive pipelines. My favorite thing is taking an "AI could do this" idea and turning it into a system that does it autonomously, reliably, millions of times.
31M+ messages ingested. 4.9K+ autonomous agent runs. Still counting.



