Senior Agentic (AI) Engineer

Worth AI is hiring a Senior Agentic AI Engineer to design and ship production agent systems that automate KYB, underwriting, and risk decisions on regulated financial data. You’ll own agents end-to-end architecture, retrieval, tools, evals, and production deployment and partner closely with our Chief AI Officer, applied scientists, and platform teams. Responsibilities - Design and ship multi-step agentic systems (planner/executor, tool-using, multi-agent, human-in-the-loop) for onboarding, underwriting, case review, and continuous monitoring. - Architect agent graphs in LangGraph (or comparable — CrewAI, AutoGen, Claude Agent SDK) with explicit state, durable execution, retries, and safe fallbacks. - Build the retrieval layer powering our agents — chunking, hybrid search, reranking, and grounded citation. - Own the eval stack: golden sets, offline regression suites, LLM-as-judge, online A/B and shadow evals, and red-teaming for jailbreaks, prompt injection, and PII leakage. - Expose agents to production systems via well-typed tools and MCP servers. Treat tool surface area as a product. - Drive production MLOps: deployment, versioning, traffic shaping, cost/latency budgets, tracing, and on-call playbooks for agent incidents. - Partner with security and compliance to keep agents inside SOC 2, GDPR, CCPA, and fair-lending posture — auditability and explainability built in, not bolted on. - Mentor engineers on agent patterns, prompt hygiene, eval discipline, and LLM failure modes. - Technology Stack - Languages: Python, Node.js, TypeScript - Agent / LLM frameworks: LangGraph, LangChain, Claude Agent SDK, MCP, OpenAI SDK - Models: Anthropic Claude, OpenAI, open-weight where appropriate - Retrieval & Data: PostgreSQL, pgvector, OpenSearch, Kafka, Redshift, Redis - Infra: AWS, Kubernetes (EKS), ArgoCD, Terraform - Evals & Observability: LangSmith / Langfuse / Braintrust-style tooling, DataDog Requirements - 5+ years of software engineering experience, with 2+ years building production LLM or agentic systems (not just notebooks or demos). - Hands-on experience with a modern agent framework (LangGraph strongly preferred) and a track record of shipping agents that run, fail gracefully, and recover. - Strong RAG fundamentals chunking, embeddings, hybrid retrieval, reranking, grounding — and judgment about when RAG isn’t the right answer. - Real eval experience golden sets, offline and online evaluations, used to make ship/no-ship calls. - Production MLOps fluency: deployed LLM workloads under real latency, cost, and reliability constraints. - Strong Python; comfortable in TypeScript / Node.js. - Solid systems engineering instincts APIs, async patterns, queues, databases, distributed system failure modes. - Calibrated communicator; thrives in ambiguous, fast-moving environments. - Prior experience in fintech, lending, payments, KYB/KYC, fraud, or AML. - Experience building MCP servers or other structured tool interfaces for LLMs. - Background in classical ML (ranking, scoring, calibration). - Experience designing explainable / auditable AI workflows for regulated environments. - Open-source contributions to agent frameworks, eval tooling, or retrieval libraries. - AWS depth (EKS, MSK, RDS, S3, Lambda) and IaC with Terraform. Success Metrics - Agent Quality: Measurable improvements in task success rate, grounding accuracy, and hallucination rate on our eval suites. - Production Reliability: Agents you own meet defined SLOs for latency (P90/P99), tool-call success, and cost per task. - Velocity: New agent capabilities go from prototype to production in weeks, without skipping evals or guardrails. - Risk Posture: Zero material incidents tied to prompt injection, PII leakage, or unsafe tool use on agents you own. - Force Multiplier: Patterns, tools, and eval scaffolding you build get adopted across engineering. All Remote Hires will be required to travel to Orlando, Florida at least twice per year for