AI-enabled software engineer / reliability
An AI-enabled software engineer with deep expertise in reliability — five years keeping large systems inside their SLOs, on-call for the ones that matter. I build agent-first: the agents root-cause their own breakages, shepherd their own incidents, and write their own reverts.
how I work
Reliability used to be measured in how fast a human could be woken up. I think the better question is what should never need a page at all. So I design systems where agents hold the pager first: they investigate, they attribute, they revert, and they hand a human the one decision that actually needed judgment. Same SLOs — far fewer humans losing sleep over them.
Five years of keeping production quiet.
Production Engineer — Recommendation System ML
DevOps / Infrastructure Engineer — R&D Engineer II
B.S. Computer Science — double major in Biology
Shipped software and production case studies.
Per-app volume control for macOS: a menu-bar slider for every application currently making noise, built directly on Core Audio process taps. No Dock icon, no window — it lives in the menu bar and gets out of the way. Ad-hoc signed, installable in one curl, with CI on every commit.
An over-the-air deployment platform for robotics and IoT — think Vercel, but the target is a fleet of devices in the field. Infrastructure as code on Pulumi and GCP, a FastAPI control plane, Pub/Sub for rollout orchestration, BigQuery for telemetry, and a Go agent on the device itself. I ran pilots with R&D labs and small businesses, rolling out and monitoring across very different environments.
Infrastructure and frontend for an online hospitality booking DApp on Next.js, Solidity and Cloudflare CDN. I defined and integrated the Ethereum smart contracts behind its token-based transactions, and tuned the booking flow for performance. The token’s market cap passed $46,000 as of Sep 2024.
A relogger: it picks up logs already being written across the fleet, re-emits them as structured metrics into InfluxDB, and surfaces them through Grafana. Nothing upstream had to be rewritten to be observable — and incident alert latency fell to roughly five minutes across three product families.
A distributed MQTT broker — MQTT 3.1.1 and 5, clustering with replicated queues, rules and connectors, and an admin interface, all in one binary. MIT licensed throughout: no enterprise edition and no open-core line.
A drop-in React component for web3 wallet connection — the boring, fiddly part of every DApp, packaged once and reused.
A lightweight, distributed secret manager you run in your own cloud — AWS, GCP, a VPS, Kubernetes, anywhere that runs a container. It brings its own database and issues its own tokens, so there is nothing to register and nothing external to connect.
What I reach for.
Open to conversations about reliability, production engineering, developer infrastructure, and anything that makes a system easier to trust.