ML Engineer/Researcher Engineering London

Join us to define how good AI is with people’s money, and build the models that do it better. Financial advice has been too expensive, exclusive, and hard to access for too long. At Clove, we combine trusted advisers with technology to close the advice gap and help people make better decisions with their money. AI can help, but there is a clear limitation today. Many people already use AI for personal financial guidance (over 50% of people), yet there is still not enough focus on measuring how models perform on the practical decisions that matter in real life. When you move beyond simple questions into real analysis, performance drops. Numbers are fragile, and the layered rules that govern real advice (tax, product constraints, suitability, edge cases) create long chains where small errors compound. In practice, even frontier models can miscalculate, miss constraints, or become unreliable on longer documents.

We plan to address this in two parts. First, we will build rigorous, public benchmarks that reflect real personal-finance tasks and make model strengths and failure modes clear. Second, we will train our own models on top of open source, tuned for financial advice and evaluated against standards we care about: numerical correctness, rule-following, and robustness on long, messy inputs. Our ambition is to be an authoritative voice on how AI is changing how people interact with money, grounded in measurement and backed by systems that are safe to use in a regulated product. As one of our first ML hires, you will own this work end to end.

What you’ll work on:

• Benchmarks: Define and build evals for personal-finance guidance and advice, including numerical accuracy, constraint satisfaction, and long-document robustness.
• Model building: Fine-tune, distill, and post-train open models into Clove models for financial advice.
• Research: Investigate why models fail on numbers and layered rules, and turn those findings into methods that improve reliability and auditability.
• Into the product: Ship what you build into our adviser platform and consumer app, where correctness matters.

We are a small, senior team. The product is a modular Go monolith with React frontends on GCP, and you will have room to shape the ML stack.

Your background:

• Strong ML fundamentals. You understand modern LLM training and post-training, and can reason about model behaviour from first principles.
• Rigorous about evaluation. You care about measurement quality and can design tests that are hard to game.
• You ship. You can take an ambiguous research question from idea to trained model, evaluated system, and deployed artefact.
• Ownership. You take end-to-end responsibility and bring others along as you execute.
• Product-minded. You care about the user problem and can balance research depth with shipping.
• High standards. You prioritise correctness, especially where 'roughly right' is not acceptable.

Nice-to-haves:

• Experience training or fine-tuning open-source models (LoRA/QLoRA, full fine-tunes) and operating the supporting infrastructure.
• Experience with RL post-training (DPO, GRPO, PPO) or strong opinions on when to use it.
• Work on numerical, tabular, long-context reasoning, tool use, or retrieval where correctness is non-negotiable.
• Experience in regulated or high-trust domains where auditability and being right matter.
• Comfort working in the open, including published work, open benchmarks, or open-source contributions.

Your impact:

• Establish benchmarks that set a clear standard for how AI performs on personal-finance tasks.
• Deliver Clove models that are measurably more reliable on numbers, constraints, and long documents.
• Improve the safety and auditability of AI in a regulated advice workflow.
• Help define how we evaluate, train, and decide when a model is ready for real use.

Our offer: A chance to join a small, senior team building Clove from the ground up, with real ownership and a direct line from research to production impact.

Our process: A small number of conversations with the founding team, plus a practical exercise if it helps, and clear expectations on both sides.