About the role
CloudCrane turns a customer's messy catalog into data an AI agent can be trusted with: every value checked against a contract the customer wrote, every value carrying a receipt for how it was decided, and a served tool that cannot return a record it isn't sure is safe.
This role owns whole slices of that, not tickets. You will pick up a capability at the point where it is still a sentence ("a customer should be able to hold a value back from what is served") and carry it through the schema, the API, the engine, the dashboard and the tests until it is something a customer can use and an auditor can check.
The hardest part of the job is not the AI. It is making a distributed pipeline behave deterministically: runs that resume where they stopped, ranking that returns the same order twice, and an exclusion rule that holds even when a value is missing. Past review passes here have caught a zip-bomb bypass, a double-commit race and non-deterministic ranking, all after the feature 'worked'.
What you'd do
- Design and build slices that run all the way through: import, contract, enrichment run, curation, release, served tool
- Keep safety exclusions enforced in the query rather than in a prompt, and prove it with tests that fail when the rule is removed
- Make ranking behave: structure-first embeddings, anchor gating, deterministic tie-breaks and checks that run before a release is published
- Review other people's work for correctness, race conditions and the failure modes a passing test suite misses
- Keep the vocabulary straight: one name per concept across the schema, the API and the app, because renames get expensive fast
What we need
- Five or more years building production backends, with real ownership of something that had users
- Strong SQL: you can read a query plan, and you know why a filter belongs in the database rather than in application code
- TypeScript at depth, or another statically typed language you would pick this up from quickly
- You write tests as part of building, not after, and you can explain what each one would catch
- You have debugged a race condition in production and can describe how you proved the fix
Nice to have, not required
- Vector search, embeddings, or retrieval systems where precision mattered more than recall
- Experience with append-only or audit-trail data models
- You have built something an external developer had to integrate against
How we work
- Small team, thin slices, shipped weekly. Nothing sits on a branch for a month.
- Reviews are real. Every feature ships with its tests, and a bug found in review is cheaper than one found by a customer.
- Safety-critical means the boring answer usually wins: fail closed, keep the receipt, don't guess.
- We only claim what ships. That applies to the product, the roadmap and the offer letter.
How hiring works
- 1A 30-minute call about something you've built and what was hard about it.
- 2A working session on a real problem from this codebase. You can drive, or we can pair.
- 3A conversation about how you make decisions when the evidence is thin.
- 4References, then an offer. We aim to answer within a week at every stage.