We’re looking for an AI Research Engineer based in London to shape the next generation of Tunic Pay’s models, to determine a scam with multimodal unstructured data within the latency of the payments system. If you're excited about making the world a safer place and applying AI to the world of financial systems, read on to learn more ⬇️

About the role 🎭

You’ll work with the founders, product and engineering to make advanced AI Tunic Pay’s moat. You’ll own the models and the pipelines around them, and uplevel how the whole engineering team builds with AI. Early projects are likely to include:

Our own small language model. Fine-tune an open model on our fraud labels and run it inside the bank’s tenant and eventually cross-tenant.

Embeddings. Represent payments, counterparties and case evidence so our knowledge graph learns from every investigation.

Agentic workflows. Orchestrators, harnesses and tool-use for our investigation engine so that we can scale the resolution of alerts, alongside nuanced evals to keep looping improvements!

Enterprise reality. Tier-1 bank model governance, latency, on-prem/VM deployment, no clean data. Bank data scientists may be in the room.

Compute budget. You will have a GPU and compute budget. Happy to discuss details during the interview process.

Anything 🤷🏻‍♀️. This role will evolve quickly. Expect ambiguity, variety, and responsibility, the scope will likely change in six months.

About you 🤓

Here are some thoughts on who would be successful at the role. We know we won’t be able to assess these from your portfolio alone, but our interview process will be designed around them:

• An Auto-didact: You may spend one day studying open banking payment rails and another learning how a fraud analyst triages an alert queue. You pick up new domains fast and turn what you learn into decisions, because no amount of knowledge today substitutes for the rate of future learning.
• A Scientist who ships: You run real experiments with real baselines, and you know when “good enough, in prod” beats one more ablation.
• A Structurer: You enjoy turning an open, ambiguous brief into something the team can act on. You structure your own work without waiting to be told.
• A Force multiplier: You make the colleagues around you better at AI, through pairing & collaboration, reviews, experimentation, and the tooling you leave behind.

The hard skills we’ll value

As we look at profiles, here’s what we’ll be scanning for:

• Deep learning depth: you’ve trained or fine-tuned models (LLMs, embeddings, or both), not just prompted them. PhDs preferred / Publications welcome, but not required.