Senior Python engineers,
on contract.
Levelbrook staffs senior Python developers onto your team — for FastAPI, Django, and Flask backends, data pipelines and automation, and applied AI/LLM and vision-ML tooling that ships rather than sits in a notebook. We're generalists by temperament: Rails is our deepest track, but we work happily across Python, Rust, Go, and Node, and we ramp fast on whatever stack the problem is already in. Tested, maintainable code with the decisions written down. Corp-to-corp, MSA / SOW / NDA / COI ready on day one.
Open-source Python we wrote and tested.
Small, focused, MIT-licensed packages that fill real gaps in the Python toolchain — the kind of thing you write when you've felt the pain in production. Each is public, documented, and has a real test suite running in CI. New work, honestly labeled v0.1.0; the point is that the edge cases are right and there are tests that keep them right.
A sans-I/O incremental parser for Server-Sent Events — feed it bytes from any transport (httpx, aiohttp, sockets), get fully-parsed events. The correctness lives in the chunk boundaries: CRLF split across reads, multi-line data, BOM, incremental UTF-8. Exactly what you need to consume a streaming LLM API.
GitHub →Tiny, dependency-free retry & backoff. The trick: the backoff math is pure and seedable and time.sleep is injectable, so the AWS full / equal / decorrelated jitter strategies — and the retry decorator on top — are genuinely unit-testable without a single real sleep.
A durable invoice-intake-to-provisioning workflow engine: an async orchestrator with retries/backoff, a dead-letter queue, structured logging, SLO reporting, and pluggable ERP/ticketing connectors — the unglamorous integration plumbing done properly.
GitHub →An enterprise IAM reference implementation in pure Python: RBAC + ABAC authorization, federation/SSO claim mapping, non-human identity governance, segregation-of-duties, access certification, and a hash-chained audit log.
GitHub →The thinking behind the code.
We publish the engineering reasoning, not just the repo. Two recent deep-dives on the Python packages above:
Why "sans-I/O" (the h11 philosophy) makes a protocol parser reusable across every HTTP client, and why the whole game is getting the chunk-boundary edge cases right.
Read the essay →Exactly-once is a lie. Separate the pure backoff policy from the impure retry loop and your jitter — and your whole retry path — becomes deterministically testable.
Read the essay →We don't only write Python.
The reason to hire a generalist is that real systems are never one language. We reach for the right tool and ramp fast on an unfamiliar one — and we have public proof across several. If your stack isn't listed, that's not a blocker; it's a week.
A zero-dependency Rust crate and CLI to parse and format human-readable byte sizes ("1.5 GiB" ⇄ bytes), binary and decimal units, overflow-safe with well-defined rounding. Proof we ship real Rust, not just talk about it.
GitHub →An applied vision-ML pipeline: DINOv2 features → UMAP → HDBSCAN → CLIP zero-shot labels → a React viewer, clustering 20k+ photos by viewpoint. End-to-end applied AI, deployed.
GitHub →A Ruby encoder for the Vercel AI SDK Data Stream Protocol — drive useChat/useObject from a Rails/Rack backend. Same streaming-LLM problem space as sansio-sse, a different language.
What we staff.
Senior contract engineers across the Python landscape — and the adjacent work that real Python systems always drag in.
Need Python hands this week?
Tell us what you're running into — a FastAPI service to build, a data pipeline that keeps falling over, an LLM feature to get into production, or just an extra senior engineer who can pick up your stack quickly. No pitch; we'd rather understand the problem.
Email the Python practice →Common questions.
Is Python your main thing, or are you really a generalist?
Both, honestly. Ruby on Rails is our deepest production track, and Python is a real, used-in-anger second language — FastAPI/Django backends, data pipelines, and applied AI/ML tooling, with open-source Python packages to show for it. We're generalists by choice and ramp fast on a new stack; if the work is in a language we use less day-to-day, we say so and get up to speed quickly rather than pretending otherwise.
Can you do applied AI / LLM work, or just plumbing?
Both. We've built a vision-ML clustering pipeline (DINOv2 → UMAP → HDBSCAN → CLIP) end-to-end, written streaming-LLM protocol tooling, and shipped LLM features into production apps. We're equally happy doing the unglamorous data/integration plumbing that makes any of it actually work.
How do you engage — contract, C2C, staff augmentation?
Corp-to-corp staff augmentation by default: our senior engineers embed with your team on contract. MSA, SOW, NDA, and COI ready on day one. Hourly, project, or retainer; 1099 and contract-to-hire are fine too.
How fast can you start?
Typically within one business day for scoping, and we can have an engineer contributing the same week for most engagements — including ramping into a stack that isn't on the list above.