Lead AI Engineer, Game Generation

The role

The game-generation pipeline is the product, which makes this the most important hire we're making right now. Every game in the feed comes out of it: a creator's prompt intent extraction template matching model routing asset retrieval generation validation and repair runtime play telemetry back into the loop. It's built around a fixed library of 20 proven core mechanics with hard-constraint specs (every line pass/fail checkable), and a quality score fed by real play data - completion, replay, share/remix.

You'll own that system end to end and be judged on two numbers moving at once: playability pass rate up, cost per generation down, at volume.

This is not a prompt-engineering role, and not a research role where someone else ships your work. It's also not "LLM app" building - the output here has to compile, run and be fun, millions of times, at a cost that works. You'll work with core engineering in Sofia and with game design; there's a team on the pipeline today and you become its technical owner, and lead it if you can. Comp is open - we'll pay top of market for the right person.

What you'll do

Own the full generation pipeline: intent extraction, template matching, model routing, asset retrieval, generation, validation and repair, runtime, and play telemetry feeding back into the loop.

Move playability pass rate up and cost per generation down, at volume.

Run benchmark suites, LLM-as-judge and playability pass/fail checks, and show output quality improving over time.

Own the economics: tiered model routing, RAG/templates, caching, small-model distillation.

What you bring

Have shipped text-to-game or AI game-creation tech (Rosebud, Bitmagic, Astrocade, Websim, Roblox Cube/Assistant, Hidden Door calibre) - or the next-closest thing: a prompt-to-running-app agent at scale (Replit Agent, Lovable, Bolt, v0 calibre).

Own the full loop: generation validation repair eval, with benchmark suites, LLM-as-judge and playability pass/fail checks.

Know the economics cold: tiered model routing, RAG/templates, caching, small-model distillation - have cut cost per generation 10x somewhere.

Strong enough engineer to live in the runtime too: Three.js, Phaser, Babylon, voxel engines, procedural content, asset/image generation.

Nice to have

Research credentials in generative game tech (DeepMind Genie/SIMA, Microsoft Muse, LLM-driven PCG papers).

Has led a team of 2-3.

Fine-tuned or distilled open models in production.

Startup or founding-team experience.

Logistics

Location: London, in person. Long-term this isn't flexible - we want the team in one room.

Remote until October 2026: you can start immediately from wherever you are and join the office in October. Relocation to London by October is a condition of the offer.

Full-time.

Compensation: base + equity, open - we'll pay top of market for the right person.