Machine Learning Engineer, Applied AI - Deployed

So much of the work society depends on is still slower and harder than it should be. Permits take months. Claims sit unresolved. And AI hasn't changed that — because the bottleneck isn't the models. It's the institutional context AI needs to do the work: rules, history, relationships, and judgment scattered across people, documents, and legacy systems. BrainCo exists to fix that. We build agent-native operating systems for the institutions society depends on, and our products are the first of their kind in the world — we were the first, anywhere, to fully automate construction permitting, and we're now doing the same across insurance and other industries. There is no playbook here, because no one has built this before. As a Machine Learning Engineer on Applied AI, your work begins where the demo ends: getting a model to look impressive is the easy part; making it a production decision system an institution stakes its process on is the job. The problems come in every shape — custom vision model pipelines that check blueprints against building codes at 95%+ accuracy, agents that untangle policy stacks to reveal coverage gaps, systems that predict from clinical records whether a patient is on their care path — and you'll own them end-to-end, from ambiguous customer problem to the eval that catches a whole class of errors. This is frontier ML applied where it's hardest and matters most. The problems are underspecified, the documents are brutal, the accuracy bar is institutional-grade — and the feedback loops are real, because our systems move real workflows forward every day.

Responsibilities

Turn ambiguity into shipped systems — from no problem statement, no labeled data, and no agreed definition of success, to well-posed ML problems and production deployments; Own AI systems end-to-end, including model behavior in production; Work at the research frontier with production stakes, applying LLMs, RL fine-tuning, and agentic systems; Work directly with institutions served — permit reviewers, underwriters, compliance officers — to understand how decisions actually get made; Engineer for production reality, navigating accuracy, latency, cost, and reliability in messy environments; Build eval suites and failure-mode taxonomies rigorous enough to earn institutional sign-off; Build composite AI systems and solve credit assignment across vision transformers, segmentation models, VLM reasoning, and rule engines; Build models that read dense multimodal documents such as blueprints, policy stacks, contracts, and clinical records; Design data, evals, and training loops to build and fine-tune agents on verified ground-truth outcomes; Raise the bar across the company through design reviews, internal paper club, and shared playbook

Requirements

Understanding of machine learning beyond tooling, including loss functions, generalization under distribution shift, and evaluation; Living at the bleeding edge of modern AI with instincts for squeezing the most out of LLMs and agentic systems (prompting, fine-tuning, tool use, reasoning); Ability to choose between fine-tuned segmentation models, VLMs, and rule engines and compose them into accurate systems; Treating frontier models as components to be measured, pushed, and engineered; Energized by building novel systems and comfortable when problem, data, and definition of success must be invented at once