AI Forward Deployed Engineer: AI/ML Agent Engineering.
Role & responsibilities
US Healthcare/ Retail Forward Deployed Engineer: AI/ML Agent Engineering. Genzeon Platforms, Healthcare FDE Practice. hybrid; open to remote for senior candidates; US and India delivery seats. Reports to the Healthcare / Retail FDE Practice Lead.
ROLE: Own the agent layer of product on the Aether One agentic substrate: design multi-agent graphs, build retrieval over clinical and regulatory knowledge, write the evaluation harness that gates production, and enforce the no-auto-deny and citation-chain invariants. Stay embedded with the customer through the first 90 days of production. This is an engineering seat, not a research seat.
WHAT YOU WILL OWN: Agent graph design (typed state machines, tool design, retry/fallback, LangGraph). RAG over NCDs, LCDs, payer policy, formulary criteria and coding references (CPT, ICD-10, HCPCS, LOINC, RxNorm, SNOMED). Per-criterion citation chains. Evaluation harness and quality gates (gold-standard sets, CI regression suites, drift monitoring). Deterministic guardrails and the no-auto-deny invariant. Model routing and the containment boundary (Claude and other frontier models, open-weight local models for sovereign deployments). Knowledge packs, MCP servers and Claude Skills. Customer-facing agent engineering with UM nurses, pharmacy ops and medical directors. Patent-grade specification work and ADRs.
REQUIREMENTS: 5+ years of production engineering, 2+ building LLM/ML systems in production. Production agentic systems (LangGraph, LangChain, DSPy, Temporal or custom). Strong production Python (FastAPI, async, PostgreSQL, pgvector or ChromaDB, typed interfaces, tests). Retrieval engineering depth. Evaluation as a first-class discipline. Prompt and context engineering at production scale. Healthcare regulatory literacy (CPT, ICD-10, HCPCS, NCD/LCD structure, medical necessity criteria, CMS-0057-F, PHI discipline) is a hard requirement. Fluent Claude-as-coding-partner workflow. Customer-facing maturity.
NICE TO HAVE: Fine-tuning/LoRA/distillation on clinical corpora; open-weight inference (vLLM, Ollama, llama.cpp); vision-language document pipelines; MCP server or Claude Skills authoring; sovereign/air-gapped/FedRAMP/GovCloud experience; FHIR R4, Da Vinci CRD/DTR/PAS, X12 278, NCPDP SCRIPT; open-source or research contributions; patent inventor credit; Claude API / Claude Code at production scale.
EXPECTATIONS: Complete Healthcare or Retail Brain Academy Builders Track Steps 1-4 before customer engagement. Claude-First, spec-driven development; ADRs within 24 hours of design decisions. Own the customer outcome end-to-end.
FIRST 90 DAYS: Day 1-14 Builders Track and agent graph deep-dive. Day 15-45 ship first agent to non-production behind a full eval gate. Day 46-90 production ship behind a clinician-in-the-loop gate; customer-facing by day 60.