Requirements

• Staff-level applied ML/AI experience in production. You have shipped and owned real ML and AI systems used by customers, and you have set technical direction for others, not only built prototypes.
• Strong classical machine learning. Hands-on depth in time-series modeling, anomaly and outlier detection, and gradient-boosted models such as LightGBM or XGBoost, using Python, scikit-learn and pandas or Polars.
• Modern LLM and agent patterns. You have built retrieval-augmented generation (RAG), tool-using agents, and structured (schema-validated) output, and orchestrated them with LangGraph or a similar framework.
• Rigor on evaluation, grounding and hallucination control. You know how to measure whether an LLM's output is faithful to its source data, and you have built guardrails that stop unsupported statements from reaching users.
• Discipline around data leakage and honest evaluation. You design features, splits and validation so models are never trained or scored on information they would not have in real use.
• Clear thinking about what ML can and cannot claim. You can explain to non-ML colleagues where a model's output is evidence and where it is only a signal, and why that line matters when results sit next to reported financials.
• Real software engineering. You write tested, production-grade Python services, are comfortable with SQL against warehouses like Snowflake or PostgreSQL, and have shipped containerized software.
• Technical leadership. You mentor engineers and work closely with backend, data, product and evaluation colleagues to turn feedback into lasting improvements.

Nice-to-haves

• Finance or accounting domain experience. Familiarity with ERP, billing or FP&A data and the audit expectations around it.
• Feature-attribution literacy. Experience with tools like SHAP, and a clear view of where attribution explanations break down.
• Human-in-the-loop systems. You have designed review, confirmation or correction workflows and fed them back into models.
• Managed LLMs inside customer clouds. Experience running AWS Bedrock, Google Vertex AI or Azure OpenAI inside a customer's VPC so their data never leaves their environment.
• Sparse and zero-inflated data. Experience modeling rare events or data that is mostly zeros.
• Multi-deployment enterprise software. Exposure to products shipped as multi-tenant SaaS, single-tenant SaaS and on-premises across AWS, Azure and GCP.