Software engineer in New York. I design, build and operate production software, from distributed infrastructure to complete consumer apps. I use AI agents to move faster without giving up rigor.
Both are my own, built outside my day job. The architecture, specs and decisions are mine, across the infrastructure, mobile and web clients, payments and test harnesses. Agents wrote much of the code inside that structure, under a process I designed, and every change went through my review.
Live group chat that runs in episodes. A creator goes live, followers get a push, and the room fills in seconds.
A literacy platform for tutors who teach reading with the Orton-Gillingham method, an approach designed for dyslexic readers.
The architecture, the constraints and the final call are mine. Agents write code inside a structure I designed: written context, small disjoint scopes, independent review, and evidence before anything ships. It runs as project skills, so every feature takes the same six steps. For scale, Claude agents authored 323 of the 2,647 commits across both repos and co-authored 1,215 more. That measures pace, not quality.
A fresh git worktree and branch per feature, so parallel features never share a checkout.
agentA written plan in docs/plans: problem, numbered decisions, steps, tests. Grounded in the glossary and ADRs.
agentAn interview against the docs that surfaces ambiguity before any code. I answer it and make the calls.
me + agentUp to three Opus implementers on disjoint files. A producer's contract lands before its consumer starts.
agentsA separate reviewer agent reads the plan and the diff cold and ranks findings. It reports; it does not fix.
agentDraft PR. I decide what gets fixed, review, and merge. Releasing is a separate, guarded step.
meEach repo has a domain glossary and ADRs (72 in Convo, 46 in IOG), and plans are grilled against them before any code. The grill step updates them as decisions settle, so the next feature starts from the same contract.
CONTEXT.md · docs/adr/ · .claude/skills/grill-with-docsImplementers report success confidently, so a reviewer without their context checks the work. On Convo’s admin moderation PR it found that a suspended paid convo could still sell memberships. Fixed on the branch before merge.
convo · docs/reviews/pr-54-admin-shows-management.md, F1A synthetic 5,052-socket load test against production found API instances silently wedging (422 gateway timeouts). The fixes passed a proof rerun with zero 5xx, a 90-minute soak exposed the real cause, and a 2-hour soak confirmed that fix.
convo · docs/plans/bertha/13-wedged-api-instances.md, 21-log-volume-wedge.mdIn Convo, the merge skill first checks that merging cannot start a production deploy (previews and the internal admin app excepted) and stops if it can. It forbids agents from deploying; a release is a separate request from me.
convo · .claude/skills/ship/SKILL.mdConsumer, fintech and AI systems, usually the parts a business runs on: conversion, fulfillment, retirement recordkeeping, identity verification and data infrastructure. I started on the data side of a company building AI-generated music.
I design and deliver the KYC/KYB verification workflows that onboard advisors and businesses, and I’m building the system that manages advisor data through its lifecycle and propagates every change across the platform. I also lead the architecture of a greenfield AI product whose primary interface is iMessage.
Worked directly with client engineering teams to take production AI systems from implementation through deployment. I ran technical delivery across four client accounts and brought AI-assisted engineering workflows to both client and internal teams.
Architected and implemented the employment recordkeeping systems that 401(k) contribution allocation depends on. Ruby on Rails, Sidekiq, GCP.
On Conversion, I raised website sign-up conversion from 8% to 14%, was promoted to Senior in December 2021, and decomposed the monolithic conversion system into independent services. From 2023 I led a lean team on Internal Tools & Operations, including the architecture and build of a fulfillment management system that supports more than 660,000 boxes shipped a month.
Built and maintained the music data infrastructure under Amper’s internal and external products, and extended its core MySQL data model. Built tools for data management and governance, and contributed to research and deployment of machine learning models.
Data engineering consulting (later acquired by Accenture). At UPS, cloud test-automation tooling in C# and text analytics on defect data.
I’m looking for senior and staff roles where system design, production ownership and AI-native engineering matter, especially at companies building with frontier models. Email is the fastest way to reach me.
