SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Belvo is an open finance API platform democratizing access to financial services across Latin America. The company enables fintech innovators to access, interpret, and act on financial data while initiating payments securely. With $71M in funding from top-tier VCs (Founders Fund, Kaszek Ventures, Y Combinator) and 90+ team members across 20 nationalities, Belvo is solving the complex problem of connecting fintech innovators with legacy Latin American financial infrastructure.
The Intelligence Squad is Belvo's AI enabler team. This role sits at the intersection of platform-building and applied modeling: turning aggregated financial data and payments across Latin America into models that drive real decisions—credit scoring, fraud detection, transaction categorization, income verification. The bar for rigor is high because these models directly impact financial outcomes for end users and businesses.
You will be the data science reference in the squad, not executing a pre-written backlog but deciding which problems are worth modeling, defining success criteria, and raising the data science bar across the organization. You'll combine deep statistical and ML expertise with business acumen and strong engineering fundamentals.
Key Responsibilities:
- Lead design and selection of statistical and ML models (classification, regression, ranking, NLP, LLM-based) for risk scoring, fraud detection, and transaction categorization; defend choices with data.
- Use agentic coding and research tools as core workflow: exploring data, generating and testing feature hypotheses, scaffolding pipelines, writing production services.
- Partner with Product and business teams to translate ambiguous questions into well-scoped modeling problems; identify signals in Belvo's financial data that explain outcomes.
- Take models from notebook to production: build or shape pipelines, APIs, and services; own them live including monitoring, drift detection, retraining, and retirement decisions.
- Design and run experiments (A/B tests, offline evaluation) to validate that model changes move intended metrics.
- Work cross-functionally with Engineering, Data Research, Product, and Intelligence Squad to ship end-to-end; partner with business stakeholders to keep models grounded in real impact.
- Translate ambiguous business questions for non-technical audiences and executives; explain trade-offs clearly.
- Stay current with GenAI developments (model releases, papers, tooling); hold opinions on what's real vs. hype; bring useful innovations back to the squad.
- Frame problems critically: decide what's worth modeling; push back when rules, heuristics, or better data pipelines solve the problem without a model; retire models when evidence warrants.
Tech Stack:
- Python (pandas, scikit-learn, TensorFlow/PyTorch) for data science and modeling.
- SQL against cloud data warehouse for data access and feature engineering.
- Claude and Claude Code for agentic tooling (part of daily workflow).
- AWS with managed services (e.g., SageMaker).
- Experimentation platform for A/B testing and offline evaluation.
- Datadog for service observability.
- CI/CD best practices.
Requirements:
- 5+ years as a Data Scientist, Applied Scientist, or similar hands-on modeling role.
- End-to-end ownership of models in production: from problem framing to live monitoring; ability to point to measurable business impact.
- Solid grounding in statistics; can explain models at statistical level, not just API calls.
- Built and compared multiple model types (tree-based, regression, deep learning) for same problem; can justify selection.
- Proficiency in Python for data science (pandas, scikit-learn, TensorFlow or PyTorch) and SQL for data pulling and shaping.
- Deployed multiple models to production yourself (not handed off); understand engineering fundamentals: testing, version control, CI/CD, maintainable code.
- Built with LLMs beyond API calls: tool use, agents, retrieval, structured outputs; shipped at least one to production with evaluation framework (not eyeballed).
- Agentic coding tools are daily workflow, not novelty; can walk through specific task that now takes hours instead of days, explaining setup, delegation, and trust boundaries.
- Proficient in written and spoken English.
Nice-to-Have:
- Fintech, financial risk, or regulated data environment experience.
- Business/strategy background alongside data science (MBA, consulting, product/strategy role).
- Customer-facing data science role (forward-deployed engineer, solutions/applied team at data platform).
- MLOps tooling experience (model versioning, monitoring, automated retraining).
- AWS deployment experience (SageMaker or equivalent managed ML service).
- Written evals or LLM-as-judge harness for non-deterministic systems; opinions on approach limitations.
- Used LLMs in data pipeline: labeling, extraction, categorization; can articulate where they outperform classical models and where they underperform.
