Paul Shynin

I build, solve, ship, and own outcomes.

I'm a Staff Backend & Platform Engineer, most useful where a system has to change but can't stop: checkout, payments, catalog, and the platforms underneath them.

Open to Staff and Senior backend and platform roles, full-time or contract. I also take on hands-on engineering projects.

Résumé (PDF)

Changing a system that can't stopRequests from every funnel reach one stable entry point. Behind it, each funnel has its own flag, drawn as a switch: thrown one way it sends that funnel to new service, the other way it leaves it on legacy system. A kill switch linked to every flag throws every funnel back to legacy in minutes.Requests from every funnelOne stable entry pointA flag per funnelKill switchevery funnel back tolegacy, in minutesNew servicelive for migrated funnelsLegacy systemthe way back
Changing a system that can't stop. Simplified reconstruction of how I move a system off its legacy core one funnel at a time, so no single step is a bet on everything.

Outcomes

  • ~5 s, then one orchestration layer in place of per-surface integrations, to ~400 ms

    Add-to-cart latency on a checkout platform where I led up to six engineers. Cart-checkout platform

  • ~15–20/week, then ownership mapped first; no monitor without a triage guide, to ~2–3/week

    P1-P2 incidents, cut ~85% by a cross-team reliability initiative I started and led. Checkout reliability

  • Overlapping edit claim, then checked at claim time, before either agent starts, to Refused, exit 3

    AgentGrid, my coordination tool for coding agents working in parallel. Nothing is written when a claim is refused. AgentGrid

Now

Since March 2026 I've worked independently. I build developer tooling for AI coding agents, and I evaluate AI coding models on repository-level tasks.

  • AgentGrid

    Independent project, 2026

    A Go and SQLite CLI that coordinates coding agents working in parallel: agents claim files before editing, overlapping edit claims are refused, plans are flagged when the base branch moves under them, and risky diffs are scored. Version 0.1, used on my own projects.

  • agent-playbook

    Independent project, 2026

    A written method for running coding agents, kept separate from AgentGrid: six agent roles, one git worktree per slice, a fresh review session for every PR, and a human who approves the plan and merges.

  • AI model evaluation

    Contract work

    Evaluating AI coding models on repository-level tasks: applying generated patches, running automated tests, and scoring correctness and engineering quality.

  • Backend and data engineering

    Independent project

    A scheduled Python data pipeline on OCI with PostgreSQL, scoring, alerting, and a dashboard.

About

Paul Shynin
  • Make change reversible. One stable entry point, a flag per funnel, and a kill switch, as drawn at the top of this page.
  • Decide what not to build. Help choose a platform instead of rebuilding one, build in-house when the system will be permanent, and retire a service by moving its consumers off first.
  • Lead through influence, stay hands-on. I led up to six engineers on a checkout platform and coordinated four partner teams, and I still write code alongside the people I lead.

Experience

Staff describes the scope I work at and the roles I'm seeking. At Peloton my title was Senior Software Engineer; I've worked independently since March 2026.

  • Independent Projects and Contract WorkMar 2026 – Present

    Backend, Platform & AI Engineering

  • Peloton InteractiveNov 2022 – Feb 2026

    Senior Software Engineer, Commerce Platform

  • Peloton InteractiveSep 2019 – Nov 2022

    Software Engineer, Commerce

  • SophosJan 2019 – Sep 2019

    Senior Software Engineer, Email Security

  • Hitachi VantaraMar 2017 – Jan 2019

    Software Engineer, Data Platform

  • AG MednetFeb 2016 – Mar 2017

    Software Engineer, Healthcare

Hiring for backend or platform work, or need a system changed without stopping it? Get in touch.