Back to jobsArchitect end to end solutions: data ingestion → retrieval (hybrid + rerank) → RAG/GraphRAG → orchestration → inference → observability; design for latency, cost, accuracy.
Design multi agent systems (hierarchical/peer to peer/ReAct) with shared memory (short term state, vector DB, knowledge graph), tool use, planning, and guardrails.
Productionize on NVIDIA AI Enterprise: package models as NIM microservices, optimize with Triton/TensorRT, track with LLMOps (evals, canary, drift).
Implement Responsible AI: safety (hallucination mitigation, jailbreak detection), privacy/PII, bias testing, lineage/citations; align to NIST AI RMF / ISO 42001 where relevant.
Integrate with enterprise platforms (M365/Copilot, CRM/ERP/ITSM), lakehouse (Snowflake/Databricks), and vector stores (pgvector, Pinecone, Milvus).
Deliverables: reference architectures, security blueprints, Bill of Materials (SBOM), VEX, deployment runbooks, cost models/TCO.
Mentor engineers (Python/PyTorch), enable squads on LangChain/LangGraph, retrieval evaluation, and AgentOps/LLMOps.
