• Own end-to-end development of LLM features: problem framing, data prep, prototyping, offline/online evaluation, deployment, and monitoring. • Build retrieval-augmented generation (RAG) pipelines with vector search (e.g., FAISS, Pinecone, OpenSearch/KNN) and document orchestration. • Implement prompt strategies, tool use/function calling, and guardrails for safety, bias, and privacy. • Integrate models in production services (REST/GraphQL/gRPC), including auth, rate limiting, and observability. • Stand up evals and experiment frameworks (A/B tests, golden sets, regression suites) with clear success metrics. • Optimize for latency, cost, and quality: prompt compression, caching, model selection, fine-tuning/LoRA, distillation where appropriate. • Collaborate with DevOps/MLOps/Platform to automate CI/CD, data/version management, and feature flags. • Embed with CX/Support to mine tickets, chats, and call transcripts; convert VOC into training/eval datasets and backlog priorities. • Instrument user journeys and define online/offline evals (win rate, hallucination rate, TTR, CSAT/NPS); run A/B tests and ship iterative improvements. • Build feedback loops (thumbs-up/down, rationale capture, escalation) and human-in-the-loop fallbacks that protect quality. • Own reliability and UX details that matter for customers: latency budgets, safe fallbacks, clear handoff to human agents, accessibility. • Partner with Trust/Legal/Security to ensure privacy-by-design and compliant data handling; implement guardrails and red-team mitigations.