I run Levelbrook Consulting, a Rails-first practice on the US west coast. Ten years shipping production backends, the last several as senior engineer on a multi-tenant B2B platform that grew from roughly $4M to $13M ARR. Available now for contract, freelance or C2C, flexible on rate, and happy to start small so you can see the work first.
Recent work I can talk about in detail:
* Cut $444k out of a Glacier bill covering 2.6 PB of objects.
* Rearchitecting the upload pipeline behind about 3 PB of ingest onto Elixir, Phoenix and Ecto.
* Deployed DINOv3 on Kubernetes for vehicle damage comparison: embed each new photo, cosine-similarity it against that vehicle's own history, surface the prior shot from the same angle. Turns "is this damage new?" into a diff.
* Built an MCP server so support could run data entry through Claude as a review queue, approving or rejecting proposed changes the way an engineer reviews a pull request.
* Tripled my year over year PR output with agentic development: Claude Code, Greptile, GitHub Actions, Playwright. On a large real codebase, not a toy.
Pushing deliberately into Go and Python. I was at GopherCon this year and published two original Go tools:
* tether (https://github.com/tachyurgy/tether), static goroutine leak analyzer on x/tools/go/analysis. Benchmarked on the whole Go 1.25 stdlib; found a real WaitGroup hang in database/sql's tests.
* weave (https://github.com/tachyurgy/weave), deterministic interleaving explorer, roughly what Loom is for Java. Stateless DFS with replay, vector clock race detection, delta debugging.
Rails is where I am fastest but I have shipped Python and JS for years and I am not precious about the stack. Backlog nobody has time for, a migration that keeps slipping, a pipeline that needs an owner: email me.
Senior full-stack engineer, nine years shipping production web software, principal at Levelbrook Consulting. Deep Rails plus a lot of recent TypeScript and Go. I like owning things end to end: product feature to schema to deploy to the monitoring that proves it works.
Recent public work, all live:
https://redraft.levelbrook.com (fiction revision studio: three LLM editorial passes stream side by side, word-level diff, per-change accept/reject backed by a property-tested merge; TypeScript/React/TipTap/Express/Postgres)
https://demo.levelbrook.com (Rails 8 + Hotwire workspace: Turbo 8 morphing Kanban, live LLM token streaming over SSE)
https://github.com/tachyurgy (more, including a Go exactly-once ingest gateway and a Python campaign-delivery service, both with real concurrency tests)
Open to senior IC full-time (remote, US) or contract engagements through Levelbrook.
Remote: Yes, remote only
Willing to relocate: No
Technologies: Ruby on Rails, PostgreSQL, Sidekiq, Hotwire, Python, TypeScript/React, Go, Elixir/Phoenix/Ecto, AWS, Kubernetes, Playwright
Resume: https://levelbrook-resume.s3.amazonaws.com/levelbrook-consul...
Email: levelbrookteam@gmail.com
I run Levelbrook Consulting, a Rails-first practice on the US west coast. Ten years shipping production backends, the last several as senior engineer on a multi-tenant B2B platform that grew from roughly $4M to $13M ARR. Available now for contract, freelance or C2C, flexible on rate, and happy to start small so you can see the work first.
Recent work I can talk about in detail:
* Cut $444k out of a Glacier bill covering 2.6 PB of objects.
* Rearchitecting the upload pipeline behind about 3 PB of ingest onto Elixir, Phoenix and Ecto.
* Deployed DINOv3 on Kubernetes for vehicle damage comparison: embed each new photo, cosine-similarity it against that vehicle's own history, surface the prior shot from the same angle. Turns "is this damage new?" into a diff.
* Built an MCP server so support could run data entry through Claude as a review queue, approving or rejecting proposed changes the way an engineer reviews a pull request.
* Tripled my year over year PR output with agentic development: Claude Code, Greptile, GitHub Actions, Playwright. On a large real codebase, not a toy.
Pushing deliberately into Go and Python. I was at GopherCon this year and published two original Go tools:
* tether (https://github.com/tachyurgy/tether), static goroutine leak analyzer on x/tools/go/analysis. Benchmarked on the whole Go 1.25 stdlib; found a real WaitGroup hang in database/sql's tests.
* weave (https://github.com/tachyurgy/weave), deterministic interleaving explorer, roughly what Loom is for Java. Stateless DFS with replay, vector clock race detection, delta debugging.
Also: https://demo.levelbrook.com (Rails 8 + Turbo 8 morphing, SSE LLM streaming), ai_stream on RubyGems (Ruby impl of the Vercel AI SDK stream protocol), https://lingua.levelbrook.com.
Rails is where I am fastest but I have shipped Python and JS for years and I am not precious about the stack. Backlog nobody has time for, a migration that keeps slipping, a pipeline that needs an owner: email me.