AI Engineer: Python / FastAPI / RAG

About The Role
We’re hiring an AI Engineer to own the AI layer of a subscription SaaS product in the financial space a retrieval-based assistant, the safety layer that keeps it honest, and the analytics engine behind it.
The product is live with paying users. This is product engineering, not research. The work is making AI output correct, cheap, fast, and reproducible good retrieval, tight safety controls, deterministic maths, and evidence that a change actually improved things.
The stack is Python microservices on FastAPI, with PostgreSQL, Docker, and AWS.
We are an AI-first engineering team Claude Code, Codex and Cursor are part of how we ship, not an experiment on the side.
You’ll work on a small team, take a feature from a spec conversation to production yourself, and talk directly to client-side engineers while doing it. We need someone who reads an existing codebase, takes ownership, and drives decisions not someone who waits for the next instruction.
What You’ll Work On
Retrieval quality end to end query classification, chunking and embedding strategy, vector search, context management
Content ingestion pipelines that are durable and observable, so a document failing halfway is visible and recoverable
Evaluation harnesses and regression sets, so every prompt and model change is measured rather than judged by eye
Tracing and observability for the AI layer knowing what a model was asked, what it answered, and what it cost
Safety and moderation keeping the assistant on-topic and unable to expose what it shouldn’t
Handling personal and sensitive data correctly before it ever reaches a model
Keeping LLMs out of arithmetic anything numeric or auditable runs as deterministic code, and the model narrates the result rather than computing it
Model routing and cost control picking the cheapest model that still passes evals, and proving it did
Integrating third-party data APIs that feed the AI and analytics layers
Shipping into FastAPI services on AWS with database migrations, structured logging, and tests that run in CI
Requirements
4–6 years of professional Python backend experience, FastAPI strongly preferred
Production experience with LLM APIs shipped to real users, not notebooks or tutorials
Built RAG systems in production embeddings, chunking, vector databases, retrieval evaluation
You already build with AI coding tools every day Claude Code, Codex, Cursor or similar. This is how our whole team works, so we expect you to arrive fluent in them, not learn them here
You can prove a prompt or model change made things better evals, benchmarks, before-and-after numbers. Experience with LLM tracing and observability tooling transfers directly
Judgement about when not to use an LLM. Anything numeric or auditable belongs in deterministic code
Solid PostgreSQL schema design, migrations, query optimisation
Async Python you can defend retries, timeouts, and sane failure handling around every external call
You write tests for AI code too, and you instrument what you ship
Comfortable with Docker and AWS
Strong English, written and spoken. You will be in direct contact with international clients and their engineering teams
You give and take code review without friction, and you raise blockers in writing early instead of going quiet
Comfortable owning a codebase you didn’t write and improving it without hand-holding
Preferred
Workflow orchestration engines for long-running or multi-step pipelines
MCP servers, custom agents, or your own AI tooling built on top of Claude Code or Cursor
Fintech or financial data products
PII, compliance, or data-governance work
Fine-tuning, distillation, or smaller-model routing to cut inference cost
WebSocket or real-time systems experience
Worked with international teams across time zones
Working With Us
We believe good work doesn’t require burning people out. We respect boundaries, trust the people we hire, and don’t chase anyone after hours. What we do expect is ownership when something is yours, you care about it. Decisions live in tickets and PR descriptions, not in one person’s head.

Work arrangement
No