AI EngineerUp to £250,000
About the Role
CATCHES builds physics-backed AI for garment simulation and virtual try-on, live with AMIRI and Fanatics and launched publicly at NVIDIA GTC 2026 after several years in stealth. They're now building a new consumer product on top of that foundation: a personal, proactive AI assistant that knows how clothes fit you, grounded in CATCHES' own sizing and virtual try-on data rather than a generic model's guess. You'll take the V1 from prototype to production, working directly with the founders, and own the agent end to end: the loop, memory, context, tools and the eval suite behind it.
You'll also shape how the agent shows up to the user and make the call on when a new model, agent pattern or tool earns its place on quality, reliability, latency or cost. Evals are half the job — you own the outcome numbers behind them: did the user find something that fits, did it get returned, what did the conversation cost. The stack isn't fixed; you help choose it.
What They're Looking For
- You've shipped LLM-powered features to real users, ideally consumers, and can point to a production agent you built end to end, plus at least one other product or substantial system you built from scratch
- You're fluent in harness engineering — context compilation and token budgets, tool gateways and MCP, compaction and durable state, tracing and recovery — and you've weighed owning a harness against renting a managed runtime like the OpenAI Agents API, Claude Agent SDK or LangGraph
- You understand agent memory in depth, from file-based versus database-backed stores to hybrid retrieval, versioning, and forgetting and decay, and you treat memory hygiene and user consent over what's remembered as first-class concerns rather than afterthoughts
- You bring strong product engineering in TypeScript or Python, you raise evals and failure modes before anyone asks, you make architectural decisions independently without adding complexity, and you use AI coding tools daily
- You treat security as the default rather than a checklist item, since this product holds account access and purchase authority on the user's behalf
- You don't need a PhD, publications or a research background — CATCHES cares about what you've shipped, not your credentials
About CATCHES
CATCHES is physics-backed AI for garment simulation and virtual try-on, live with AMIRI and Fanatics and launched at NVIDIA GTC 2026 after several years in stealth. Investors include Antoine Arnault, Natalia Vodianova, Roy Chung (Apollo.io), Dillon Erb (Paperspace) and Sarah Willersdorf (former Head of Luxury, BCG). It's a small, experienced team that values shipping, experimentation and autonomy, with an early say in product and architecture for whoever joins.
Questions about this role? Just ask Dex.
Talk to Dex to go deeper on this role, and it'll surface the others that fit you. When it's a match, you get a direct intro to the hiring manager. Ask anything before you commit: the real comp, the team, what the hiring manager actually prioritises.
- 1
Tell Dex what you want
Ten minutes on a call. Dex learns what great work looks like for you.
- 2
React to refine matches
Save, apply, or pass on roles. Dex tunes the matches to your taste over time.
- 3
Land the role
We represent you to the hiring team and handle introductions, scheduling, and offers.
Dex gets to know what you're great at and what you want next, then surfaces roles at companies like Catches.