Senior Machine Learning Engineer

“I can succeed as a Machine Learning Engineer at Capital Group” 

We are looking for someone who can take a vague question from an investment professional, find a real answer in messy data, prove the answer holds, and build the thing that delivers it. 

You will join the AI Insights team. We build the insight layer on top of Capital Group’s investment data: multi-agent systems that answer investment questions with citations, the evaluation methods that tell us whether those answers are any good, agents that take on expert analyst workflows end-to-end, and the extraction work that turns unstructured research, calls, and filings into reusable insight. Some problems here are better served by a conventional supervised model, and part of the job is knowing which is which. 

Our work reaches across the investment organization, from research analysts to governance specialists to the teams behind portfolio and order management. Each partner brings its own data, its own workflow, and its own idea of what a good answer looks like. You go deep with one rather than skim, and you end up learning parts of the business most engineers never see. 

Whether systems like these actually work, and how anyone would know, is still an open problem in this field, and making it answerable here is a large part of this role. This is applied science with a delivery bar, not a research lab: the answers have to hold up to people making real investment decisions, and they have to arrive as something working rather than a paper. Everyone on this team builds. There is no version of this role where you hand a design to someone else and review what comes back. We work hand in hand with a partner engineering team that owns the platform, so your time goes to the insight and the evaluation rather than the infrastructure underneath it. 

“I am the person Capital Group is looking for.” 

You will: 

• Sharpen an underspecified ask into a problem worth solving: what is really being asked, what would count as an answer, what evidence would settle it. 

• Pull signal out of messy, incomplete data, and tell a real result from leakage, a lucky split, or a metric that flatters itself. 

• Design the evaluations that tell us whether a Generative AI system is working: eval sets, success criteria, LLM-as-judge and its failure modes, and the judgment to know when a number measures what you think it does. 

• Run the experiment that settles the question the team is arguing about, and write it up so the decision is reproducible, including the criteria you committed to before you saw the numbers. 

• Design and build agent systems that produce insight. Decompose the task, choose the orchestration, decide where a human belongs in the loop, and recognize when a single model call or a simple deterministic step is the more honest answer. 

• Build your own prototypes end-to-end, using AI coding tools to move fast while keeping the output clean and working. 

• Take your projects from a rough idea to something people use, starting with a short design you shape together with the team. 

• Strengthen the team’s craft through design and code review, and by mentoring on experimental design and rigor. 

Required Experience 

These are what we weight most heavily: 

• Research depth and scientific rigor. A track record of extracting real signal from messy, ambiguous data. You design clean evaluations, and you are skeptical of your own results when they look too good. 

• Abstraction and problem framing. You find the core constraint in an unfamiliar problem without handholding, and reach for a reusable structure rather than a one-off. 

• First-principles problem solving. You start from the problem and its constraints rather than a favorite tool, and reach for the simplest thing that works. 

• Applied ML and Generative AI experience in production. You have taken real problems end-to-end, from data understanding through evaluation to something people actually used. 

• AI acumen. You pick up new tools because you want to know how they work, not because someone made you. You work with AI coding assistants day to day, and can say concretely what you have built with them, where they helped, and where you had to take over. 

• Communication, collaboration, and maturity. You explain trade-offs clearly to non-technical partners, say “I don’t know” without discomfort, and state the other side of a disagreement fairly. You make the people around you better, and can get behind a direction you did not choose. 

• Ownership. You have driven ambiguous work to a result on your own. 

A floor rather than a differentiator, but we do expect it: 

• Builder judgment. 7+ years of professional experience, and still hands-on today. You can turn an idea into a working prototype yourself, read code with taste, and steer AI coding tools to a clean result rather than accepting whatever they produce. This is not about algorithmic puzzle solving. It is about building enough to make your research real. 

Preferred 

• Designing and evaluating multi-agent or tool-using systems, including a clear view of where they fail. 

• Building evaluation infrastructure: eval sets, offline and online measurement, regression and drift detection. 

• Finance or investment management, or a demonstrated ability to get fluent in an unfamiliar domain quickly. 

How we work, and what we value 

Nobody here is keeping score. Disagreement stays about the work rather than the person, and you do not have to be the loudest person in the room to have influence. A lot of the week goes to working sessions: brainstorming, design review, pair programming. The problems are genuinely ambiguous, and not all of them work out. 

Rigor. You try to break your own result before anyone else does. Ownership. You are a driver, not a passenger. Humility. A better argument can change your mind. Pragmatism. You know when a rough answer is enough and when it has to be airtight.