Levelbrook is a US-based senior Ruby on Rails practice for teams that need a drop-in senior engineer on contract. Rails, Hotwire, PostgreSQL and AWS are the core — product features, APIs, integrations, performance fixes, and production hardening — with Python and Node.js where a mixed stack needs them. Billed corp-to-corp through Levelbrook LLC.
Pick the shape that matches the problem. Rails is the core of every engagement — with Python and Node available where a mixed stack calls for them — and the work is scoped around the outcome, not the hours.
A drop-in senior seat on your engineering team. We pick up tickets, ship features, review PRs, and carry pager weight — same as a full-time senior, on a contract you can end any time.
Slow queries, noisy-neighbor problems, a Postgres that's outgrown itself, an S3 bill that's outgrown the company. We diagnose, then fix — usually as a scoped project.
A real REST/GraphQL API for your customers, a third-party data import that has to actually work, an OAuth/SSO story that holds up to security review — Rails-first, with Python or Node where the stack calls for it.
A sample of real problems shipped on a multi-tenant B2B SaaS in the $5–15M ARR range. Each one is written up in full on the engineering blog.
A page that counted across an 800-million-row table, fixed with the right denormalization instead of a rewrite.
Read the write-up → Rails · PerformanceSystematically surfaced 319 N+1 query sites — ~5,940ms of wasted time — and eliminated them with proper eager loading.
Read the write-up → AWS · FinOpsCut a petabyte-scale S3 line item roughly 60% — lifecycle tiering by access pattern, no user-visible regression.
Read the write-up → Integration · AWS SESA production integration with a legacy ERP, built on SES, a job queue, and compound indexes — fails loudly, recovers cleanly.
Read the write-up → Rails · GraphQLThe hidden cost of mount Admin::Engine, and a portal rebuild on React + a versioned GraphQL API.
A Salesforce sync, a read follower, and the audit that reshaped a whole credentials posture.
Read the write-up → Postgres · MigrationsLocks, batched backfills, CONCURRENTLY, and the expand/contract dual-write dance — no maintenance window.
Idempotency keys, dedup, and the dead set — designing a Sidekiq pipeline for the failures that actually happen.
Read the write-up → Mobile · SDUIA real server-driven UI runtime shipped on native iOS, native Android, React, and React Native from one source of truth.
Read the write-up → Rails · Support toolingA reparent-user workflow with a dry-run diff preview and an all-or-nothing transaction — the risky internal op, made safe.
Read the write-up → AI tooling · ProcessAn internal AI plugin system that put real, reviewable engineering skills in front of a thirty-person team — not another chat window.
Read the write-up →MIT-licensed tools we built and shipped in the open — a macOS voice agent and three dependency-free Ruby gems. Each one fills a real gap, ships a full test suite, and has a written engineering deep-dive. The bar we hold a public release to is the bar we hold client code to.
An on-device voice wake word for macOS that dispatches background coding agents — fully local until the moment of dispatch. Say "Hey Claude, <task>" and it spins up a Claude Code background agent (claude --bg); it drives any agent CLI, not just Claude.
pipx install git+https://github.com/tachyurgy/hey-claude@v0.3.0
# always-on, on-device — nothing leaves the Mac
# until an agent is actually dispatched.
wake = openWakeWord.listen() # ~0.1% CPU
speech = vad.capture(wake) # energy VAD
task = mlx_whisper.transcribe(speech) # Apple GPU, <1s
# "Hey Claude, <task>" -> background coding agent
dispatch("claude --bg", task)
A pure-Ruby backend for the Vercel AI SDK Data Stream Protocol — stream text, reasoning, tool calls, and data parts to useChat / useObject from Rails or Rack. The Ruby implementation that didn't exist.
Compiles bash globs — *, **, {a,b}, {1..3}, extglobs — into reusable Ruby Regexps. The missing picomatch / minimatch for Ruby: match S3 keys, routes, and log lines, not just files.
A faithful Ruby port of the fzy / fzf fuzzy-matching algorithm — returns a relevance score and matched positions for highlighting. Built for command palettes, quick-open, and CLI pickers.
Levelbrook is a senior Ruby on Rails consultant operating through an LLC — not a body shop with a bench it can't fill. Every engagement is led by the same senior engineer who answers your first email. When a project genuinely needs more hands, we bring in vetted senior engineers we've worked alongside for years, transparently, and the principal stays accountable for delivery.
Rails is the core practice — but plenty of Rails shops run Python alongside it, and Levelbrook ships that work directly too: FastAPI & Django backends, data pipelines & automation, and AI/LLM tooling. The proof is deployed and public — equipment-cluster, a DINOv2 → UMAP → HDBSCAN → CLIP vision-ML pipeline over 20,000+ photos, served through a React viewer.
# 20k+ rental photos -> a browsable visual
# history, clustered by machine & viewpoint.
import torch, umap
from sklearn.cluster import HDBSCAN
feats = dinov2.encode(photos) # ViT-B/14
coords = umap.UMAP(n_neighbors=15).fit_transform(feats)
labels = HDBSCAN(min_cluster_size=12).fit_predict(coords)
# zero-shot names for each cluster via CLIP
names = clip.zero_shot(centroids(coords, labels),
prompts=EQUIPMENT_VIEWS)
When a Rails engagement touches a JavaScript or TypeScript service, Levelbrook ships that too: Node.js services, REST and GraphQL APIs, OAuth/SSO, webhooks, partner integrations, worker queues, and production hardening. Same principal-led consulting model, same C2C paperwork.
export async function deliver(event) {
await withIdempotency(event.id, async () => {
const res = await partner.post(event.payload)
audit.log({ event, status: res.status })
})
}
Ruby on Rails staff augmentation, C2C consulting, APIs, integrations, and performance work — with Python and Node where they fit. Recruiters, staffing partners, direct clients, and end-clients welcome.