Head of Data & AI

The mission

Lead the team and the data and AI system behind Galaxies' Consumer Intelligence Layer: secure, permissioned enterprise data, swappable models, evals that surface failures before the client does, observability, and human judgment in the right place.

The challenge isn't running a lab or picking models. It's connecting research, data, and AI to the product and to real client decisions — with enough rigor to know when to move forward, when to cut scope, and when to recommend nothing at all.

What you'll do

• Take over a function in transition. Receive a substantive handover of systems, sources, rights, routes, evals, risks, and fallbacks, and make all of it inspectable: owners, dependencies, costs, failure modes, minimum safe mode.

• Design the enterprise data architecture: contracts, lineage, ontologies, quality, permissions, tenant boundaries, and progressive access.

• Build the learning control plane: traces, evals, model routing, observability, human review, abstention, outcomes, calibration, and cost.

• Lead Research & ML applied to product and client decisions, with baselines, comparators, falsifiers, and release criteria. Experiments that don't change a decision or produce falsifiable learning don't get core resources.

• Take at least one learning loop all the way through: observed action, outcome, cost, and learning — either retained or shut down, with an honest, reusable conclusion.

• Turn field failures into platform: primitives, evals, workflows, policies, runbooks, and reusable memory — instead of one-off fixes that don't scale.

• Build a team that decides well without you. Success here is measured by the autonomy and technical judgment of the people who work with you, not by your presence in every decision.

• Advanced English: we're preparing for international expansion, and English will be required soon.

In the first 90 days, we expect you to establish clear agreements with Product, Engineering, and Trust, make an evidence-based call between prioritizing reliability or running a first real-world test, and define a capacity plan that identifies the main bottleneck. Faced with ambiguous scenarios, we expect you to formulate testable hypotheses, weigh the impact of each choice, and set clear criteria for when to stop or redirect an initiative.

Requirements

Required:

• Enterprise data: architecture, contracts, lineage, ontologies, quality, integration, and environments with real permissioning requirements

• ML and LLMs in production: evals, RAG/memory, agents, model routing, adaptation techniques when justified

• MLOps and operations: experiment tracking, observability, drift, release/rollback, cost, latency, incident response

• Security and technical governance for data/AI, in partnership with Security, Trust, and Legal

• Applied research: experimental design, calibration, baselines, failure analysis, and translation into product

• Demonstrable judgment: you've already protected a system from overclaiming, data leakage, drift, reward hacking, hidden cost, or vendor lock-in

• English to keep up with research, vendors, models, and international collaborations

Nice to have: early-stage startup experience · enterprise context with sensitive data (CPG, financial services, retail) · having built a data/AI function from scratch · experience on a small team where leadership is hands-on

We don't require a specific degree or years of experience. We evaluate the quality of the decisions you make and the systems you can build from incomplete evidence.

How we work

100% remote, team distributed across Brazil, with an in-person offsite in São Paulo every six months · contractor (PJ) engagement

Process

Screening → Conversation with People → Technical case → Interview with CEO + Reference from the function → Offer