AI Engineer

The work is roughly half LLM agents and half custom ML, leaning slightly toward agents. Build and improve agents that find company data points on the web using search, page navigation and structured extraction tools, and that weigh conflicting sources before committing to an answer. Develop, improve and retrain the models that predict company financials such as revenue and EBITDA, and the likelihood and timing of buy and sell events. Design and maintain evaluation suites, both offline against reference sets and live in production, so we can measure quality and switch between LLM providers to balance output quality and cost. Ship your work into our production services, trace where pipelines fail, and fix the weakest links first. Design with resources in mind. Our engineering team runs the infrastructure, but you choose solutions with their compute, memory, bandwidth and model costs in view. By month six you'll know each of our agent pipelines end to end, how to evaluate them and where they most need improving. You'll also know our custom ML models well enough to improve them or retrain them on new data.

At least 3 years of building both custom ML models and LLM-based pipelines or agents, with work that ran in production rather than only in demos. Solid modelling fundamentals: feature engineering, validation that holds up on new data, and knowing when a model needs retraining. A habit of measuring before claiming. You can build an evaluation set and tell an honest improvement from a lucky one. Strong Python, including async code. Nice to have: experience with noisy or conflicting data sources and financial or private-company data. The stack Core: Python (async, fully typed with Pydantic) and FastAPI LLMs: multiple providers behind our own provider-agnostic layer, with Pydantic AI for typed agents. We add new models when they prove themselves on our evals. Web data: agents that gather data points through web search, page navigation and structured extraction tools Database: PostgreSQL with pgvector Evals and observability: Pydantic Evals and Logfire Cloud and tooling: GCP (Cloud Run, Cloud Tasks), Docker, GitHub Actions

Competitive base salary and annual bonus linked to your performance / OKRs Attractive benefits including health & wellbeing allowance, work-from-home budget, learning & coaching benefits, etc. Flexible hybrid working model with 3 days per week in our Amsterdam office Healthy work-life balance allowing for planability and personal commitments Chance to grow with the company gaining increasing responsibilities, supported by lots of coaching and a feedback-driven approach Tremendous learning and career progression opportunities International environment with hubs in New York, London, Amsterdam, Frankfurt, Warsaw, and Bangalore Culture of trust, ownership and standard of excellence and a fun working atmosphere with regular outings and events Post product-market fit and aspiring unicorn status - this is an excellent time to join & grow with us!

Gain is the Private Markets Super App, a connected global platform transforming how leading investors, advisors, and corporates source, evaluate, and execute deals. By combining investment-grade intelligence, AI-powered workflows, and proprietary data in one unified platform, Gain enables top deal teams to move faster, reduce risk, and build conviction with confidence. Gain is trusted across more than $1 trillion of private capital, used by 100% of MBB and Big Four firms, and by 80% of the world’s top 20 M&A advisory houses. Recognized for fast innovation and impact, Gain has been named both US and EU Data Provider of the Year by PE Wire, awarded Global Financial Market Review’s Best Use of AI in Finance, and ranked in Sifted’s Top 100 Fastest-Growing Companies, the Forbes 1000, and the Deloitte Technology Fast 50. Gain operates globally with offices in New York, London, Amsterdam, Frankfurt, Warsaw, and Bangalore. How we work We build AI for our customers, not for its own sake. Every project starts from a clear benefit to the people using Gain, and ideas without one don't get built, however interesting the technology. If that's how you like to work, you'll fit in well. Model choices follow the evals. We don't commit to one LLM provider; we measure, and use whichever model gives our customers the best results.