A compliance-first systems mindset — you instinctively reach for state machines, idempotency keys, exactly-once semantics, reconciliation jobs and immutable audit logs. You understand that in compliance, false negatives cost millions and false positives cost trust, and you design accordingly. Prior experience in KYC, AML, sanctions, fraud, or payments is a strong plus; experience surviving a regulatory audit is an even stronger plus
Daily, fluent use of modern AI coding tools — Claude Code, Cursor, GitHub Copilot, Codex CLI, or equivalents. You can describe the specific workflows you run, the failure modes you have hit, and the guardrails you have built around AI-generated code. AI fluency is required, not nice-to-have
Production experience integrating AI/ML model inference into Java services — REST/gRPC-based model serving, feature engineering pipelines, latency-sensitive inference optimization, graceful degradation, and drift / rollback playbooks
Hands-on experience with LLM application patterns — at least one of: RAG pipelines (vector store + retrieval + grounding) for policy / KYC / case lookup, agentic workflows (tool use, multi-step reasoning, validation loops) for analyst assistance, or LLM-based classification / extraction shipped to production
Discipline around AI output — you do not ship LLM features without evals, hallucination tests, prompt regression suites, and human-in-the-loop review for high-stakes decisions. You have an opinion on when to reach for a frontier model vs. a fine-tuned small model vs. a deterministic rule, and can defend that trade-off to a regulator
Exposure to LLM integration or AI agent frameworks is a strong plus — e.g. prompt engineering at the system level, RAG pipelines, or orchestrating AI workflows within a regulated environment
Good understanding of software engineering basics, distributed system principles, including CAP, consistency, idempotency, and exactly-once vs at-least-once semantics
Clear logic, quick thinking, and good communication skills — including the ability to write clearly for non-engineers (compliance officers, auditors, regulators)
AI Proficiency Expectations & Interview Process
AI fluency is a required engineering competency for this role, on the same level as ‘can write good Java’. We will assess it explicitly during the interview, and a strong showing here is a meaningful plus
AI-augmented workflow demo — walk us through how you use AI tools day-to-day. The prompts, the tasks you delegate to an agent vs. write yourself, how you review AI-generated diffs, and how you measure the impact (commit velocity, defect rate, test coverage). Concrete examples beat slogans
Live agentic exercise — a working session using the AI coding agent of your choice on a realistic compliance-flavoured problem. We look at how you scope, prompt, validate, and integrate AI output, not whether the agent gets it right first try
Design discussion with a probabilistic twist — expect at least one system-design question where the right answer involves deciding when not to use an LLM, and how to design the deterministic fallback. We hire engineers who can defend that boundary
Evaluation & guardrails — be ready to discuss an eval suite or guardrail you have built for an AI feature in production (e.g. an LLM-assisted SAR narrative drafting tool, an AI-assisted KYC document review step, or screening-hit disambiguation). If you have not shipped one, tell us how you would design it