The Role
We are looking for a
Senior AI Engineer to build the AI and intelligence layer - and help uphold the discipline that keeps it honest. Our output sits adjacent to externally reported financials, so "sounds plausible" is not good enough: everything the AI produces must be grounded in, and traceable to, verified data.
You'll build the machine-learning models that learn each customer's patterns, the LLM and agentic systems that produce grounded natural-language output, and help uphold the rigor that keeps that output faithful. You own significant pieces of the layer end-to-end, working closely with our Staff AI Engineer and evaluation engineer, and you build per-customer models without leaking the very signal they're meant to detect.
This is a hands-on engineering role. You turn designs into robust production systems, measure quality rigorously, help turn user feedback into durable improvements, and grow toward staff-level technical ownership.
Requirements- Build per-customer ML models that learn normal vs. anomalous behavior on time-series and operational data, using statistical baselines and gradient-boosted models (LightGBM/XGBoost) with strict anti-leakage discipline.
- Build a template-first natural-language generation layer that is strictly grounded in verified data, using guardrails and structured output to prevent hallucinated or unsupported statements.
- Implement agentic orchestration (LangGraph or equivalent) with verification and human-confirmation checkpoints.
- Integrate managed LLMs running inside customer clouds (AWS Bedrock, GCP Vertex AI, Azure OpenAI).
- Partner with an evaluation engineer to gate grounding, hallucination, and faithfulness checks in CI.
- Turn user feedback into durable model and system improvements.
- Ship production Python services with SQL/warehouse access (Snowflake, PostgreSQL) in containerized, customer-hosted deployments.
Nice-to-Have- Finance or accounting domain experience (fintech, FP&A, ERP, audit, financial close).
- Feature-attribution experience (e.g. SHAP), along with an understanding of its limits.
- Human-in-the-loop systems and feedback loops.
- Hands-on work with Bedrock, Vertex AI, or Azure OpenAI in VPC or private deployments.
- Zero-inflated or sparse-event modeling.
