Founding AI Engineer

ManaMind; Greater London; On-site; GBP 90,000 - 130,000; Full time

Benefits offered by this job

Equity

Flexible hours

High impact culture

Job summary

ManaMind is seeking a Founding AI Engineer to architect our agents' cognition, turning raw video into semantic game logic. You will own the model lifecycle, curate datasets, and fine-tune VLMs for efficient production use.

Responsibilities include building data pipelines, defining benchmarks, and delivering production-ready AI agents with a focus on low latency and scalable design. This role offers ownership, equity, and a fast-paced startup environment.

Qualifications

• Experience fine-tuning Vision-Language Models on video data.
• Ability to design data pipelines for annotated training sets.
• Experience deploying deep learning models to production with latency constraints.

Responsibilities

• Fine-tune and optimize Vision-Language Models on game footage to interpret UI, animations and game state.
• Architect model inputs to handle video sequences and temporal context.
• Build and maintain data collection pipelines from gameplay to labeled training sets.
• Create internal eval suites to measure meaningful performance metrics.
• Deliver production-ready AI agents capable of real-time gameplay.

Skills

Vision-Language Models

Video data

Model fine-tuning

Production deployment

Python

Tools

LLaVA

CLIP

Flamingo

Job description

Do you want to work on one of the hardest problems in AI, building systems that can reason and act in complex, open ended digital worlds?

Do you want your work to ship quickly and sit at the core of a product used by major game studios, with a clear and visible impact on real games?

And do you want genuine ownership, the freedom to choose the stack, design the architecture end to end, and take ideas from first principles through to production without layers of approval?

About us

We're an early-stage startup built on a simple belief: game developers should be building worlds, not chasing bugs.

We're replacing manual QA with autonomous agents that truly understand gameplay.

Our small, focused team combines deep machine learning research with strong commercial execution, driven to solve some of the hardest problems in game development.

Just over a year in, we've built category-leading technology, gained real traction, and partnered with some of the world's most storied game studios.

We're backed by top-tier investors and incubators, including SVV and EWOR.

The role

We are looking for a Founding AI Engineer to be the architect of our agents' cognition. You will be the builder who transforms raw video pixels into semantic game logic. You will own the entire model lifecycle, from curating specialized datasets to fine-tuning VLMs that run with extreme efficiency. This role demands a blend of research-grade intuition and rigorous engineering discipline. You are constructing the proprietary brain that gives our agents the agency to play.

Key responsibilities

• Train the Brain

  Fine-tune and optimize Vision-Language Models (VLMs) on custom game footage, enabling our agents to interpret complex UI, animations, and temporal game states.

• Solve for Time

  Architect model inputs to handle video sequences rather than static frames, giving agents the memory and context required to understand gameplay progression.

• Deploy for Speed

  Optimize and deploy models to run with the latency and throughput real-time gameplay demands, balancing accuracy against inference cost.

• Define the Data

  Design and automate the data collection pipelines that turn raw gameplay into high-quality, annotated training sets for specific game genres.

• Define the Benchmarks

  Create and maintain internal eval suites that measure what actually matters.

You're a good fit if you...

You have spent serious time fine tuning Vision Language Models, not just reading about them, and you are comfortable working with video data where understanding temporal context over time matters more than a single frame. You can take research ideas and turn them into systems that actually run in production, bridging the gap between experimentation and real world deployment.

Live in the Weights

You have hands-on experience training or fine-tuning multi-modal models (e.g., LLaVA, CLIP, Flamingo) from scratch or checkpoints. You know exactly what happens inside the transformer block.

Build for the Real World

You have successfully deployed deep learning models into production environments where latency and cost matter. You understand the brutal reality of trade-offs between model size and inference speed.

Thrive in Chaos

You are a self-starter who prefers the ambiguity of a zero-to-one startup environment over the structured comfort of a big tech research lab. You want your code to ship, not just be published.

Bonus

You enjoy games or complex interactive systems. You don't have to be a hardcore gamer, but you're genuinely curious about how interactive experiences are built, broken, and tested; and you like the idea of agents learning to navigate them.

What's on offer

High trust, high impact: ship real product fast, work directly with founders.

Equity

Meaningful equity ownership. You'll share in the upside you help create.

Flexible hours, and a culture built on trust and output.