Founding ML Engineer

Founding ML Engineer

Early-Stage VC Fund | London preferred (remote in Europe considered)

Compensation: Up to £200k base + carried interest

About

The fund has about 15 years of data on founders, companies, deals and talent. Right now it's spread across a CRM, a talent tool, meeting notes and calendars that don't talk to each other, so the investment, talent and ops teams each see a different slice of it.

They want to turn all of that, plus public signals like GitHub, arXiv and people leaving top labs, into something that helps them find founders earlier. Which people in their network are starting to look like future founders? Which teams have unusually strong concentrations of talent? Which weak signals are getting stronger over time?

Most of it comes down to ranking, and the answer won't always be an LLM. It might be classic ML, graph features, embeddings, learning-to-rank or a mix of all of them. Proving any of it works is a big part of the job too - it's very easy to build something that explains past winners but can't find the next one.

It's a small team and has been one of the best-performing early-stage funds out there for a long time. There's no ML team or roadmap yet, so you'd be setting the direction and building it.

You'd work directly with the partners, ideally from London so you can bounce ideas around in person, and share in the fund's carry.

What you'll do

• Build ingestion across GitHub, arXiv, Hugging Face, company registries and internal CRM data
• Merge each founder's traces into a single profile, with provenance
• Map people, companies, papers and repos as a knowledge graph
• Turn partner feedback into features and ranking, classification and anomaly-detection models
• Use bandits or RL to learn what to surface and where to look next
• Backtest the system on 15+ years of real bets and misses
• Auto-generate cited briefs on strong signals

What you'll need

• Shipped ranking, classification or time-series models on imbalanced data
• Built vector search and RAG over messy, multi-source data
• Record linkage at scale
• Reliable ingestion from APIs, crawls and bulk dumps; strong Python and SQL
• Depth productising LLMs, agents and tool calling
• Fluency in backtesting, leakage, survivorship bias and base rates
• Run a system end to end as the sole engineer

Optional Bonus

• Learning-to-rank, bandits or RL
• Neo4j or similar for modelling people and organisations
• Entity resolution across languages

Part research, part production: hypothesis, experiment, evaluation, shipped system. You'll explain trade-offs plainly to non-technical partners.

Shortlisted candidates will be contacted within 48 hours.