AI/ML Developer

About Organisation:
We are a multi-tenant operations platform an operating system for industries that run on customers, partner networks, and field operations. Six services live in production. Anchor customer is live; pre/early-revenue.TypeScript across the application services, gRPC between them, PostgreSQL with Row-Level Security for multi-tenant isolation. Python on the data and ML layer — FastAPI + SQLAlchemy + Alembic. Anthropic Claude and OpenAI APIs, pgvector for embeddings, the canonical agent SDKs. Claude Code, Cursor, and Copilot everyday — and we expect the AI/ML engineer to set the bar for how the rest of the team use them.
What you’d own
The Python data and ML layer end-to-end. Pipelines from Postgres for analytics and training. RAG over tenant documents. Agentic workflows. Embeddings retrieval. AI features that run in production with real data — not notebooks, not demos. You integrate back into the platform over gRPC.
What we look for
• Hands-on LLM API experience — production integration, not just calls
• RAG, vector databases, embeddings — and when each works vs. fails
• Strong Python — Postgres extraction, pipelines, ML training.
• Agent framework experience — LangChain, CrewAI, Claude Agent SDK, or equivalent.
• Production AI mindset — evaluation, monitoring, failure modes.
• Comfortable in TypeScript to integrate AI features over gRPCSignal.
• You’ve shipped an AI feature that runs in production, handles real data, and has measurable outcomes.
• You can describe a specific time the AI behaved unexpectedly and what you did about it — not “added a check” but designed a safeguard that holds across edge cases.
• You’ve used Claude Code, Cursor, or Copilot in a real codebase and can describe how you manage their output.
• You write clearly about technical decisions Self-selection — read these honestly before applying
• In-person, default in-office. Not remote-first.
• No level ladder. No Senior / Mid / Junior. We pay you what you’re worth, not for a title.
• Ambiguity is the job. Early-stage data is incomplete and inconsistent. You work with what exists.
• You’ll iterate. Multiple approaches before one works. That’s the job, not an obstacle.
What we offer
Comp as published — no negotiation games. Unlimited PTO. Health insurance for you and immediate family. Claude Code, Cursor, and GitHub Copilot subscriptions, fully covered. Learning and development budget.

Work arrangement
No