Transform hospitality data into an intelligent, cost-aware assistant via hybrid RAG, knowledge graphs, multi-tier memory, and agent-orchestrated workflows—all while upholding privacy, safety, and performance at Marriott.
Requirements
- Proven success fine-tuning open-weight models (Llama-2/3, Mixtral, Phi-3, etc.) to production.
- Deep grasp of KG query languages (Cypher, SPARQL) and embedding-graph fusion.
- Documented cost-governance wins (quantisation, distillation, traffic shaping).
- Hands-on agent orchestration (Autogen, CrewAI, LangGraph, or custom MCP).
- Security first mindset; experience sanitising prompt logs for PII/bias.
- Experience deploying multimodal or voice interfaces at scale.
Responsibilities
- Build hybrid RAG + KG retrievers with regional embeddings.
- Implement session + encrypted long-term guest memory under retention rules.
- Design model-routing & prompt-caching with real-time cost/latency dashboards.
- Embed guardrails in CI/CD; crypto-sign evidence into the Studio’s self-cert repo.
- Lead engineers and autonomous agents; set prompt standards, vet agent plans, and deliver measurable NPS/RevPAR gains.
Other
- This is a temporary position.
- 8+ years of experience in AI and software development
- Published benchmarks or blogs on LLM optimisation.
- Washington Applicants Only
- The application deadline for this position is 21 days after the date of this posting, August 26, 2025.
