[DT] : AI Solution Architect / Agentic Workflow Lead

Location: Noida, India; Remote, India

Department: Software Delivery

Experience: 8+ Years

• Complex RAG / grounded retrieval over messy enterprise documents — chunking, embeddings, hybrid retrieval, reranking, citations, metadata, and access-control-aware retrieval, plus sound judgement on when to retrieve versus rely on long-context strategies, and how to manage context across long-running agent sessions
• Agentic workflow orchestration using a stateful framework (LangGraph, Semantic Kernel, Durable Functions, Temporal or equivalent), with human-in-the-loop, safe tool use, error handling, tracing, and working knowledge of emerging agent interoperability standards (MCP, A2A)
• Modelling business processes into deterministic rules, decision trees, and workflow states (in our case, a bid-writing workflow), with reliable structured/schema-conformant output generation to keep those states trustworthy
• Securing agentic pipelines against prompt injection and other adversarial inputs, particularly where third-party documents (tender packs, uploaded content) feed directly into tool-using agents
• Evaluation and LLMOps. Automated eval pipelines for correctness, groundedness, citation accuracy, latency, cost and safety, plus prompt/version management and regression testing. We currently use langfuse for tracing but would like to expand into evaluation.
• Model routing and reasoning-effort strategy — matching model choice and reasoning budget to task criticality, and applying cost levers like fine-tuning or distillation for narrow sub-tasks where appropriate
• Azure-first delivery (Azure OpenAI / AI Foundry, Azure AI Search, Functions or Container Apps, Key Vault, Entra ID) with a strong grasp of tenant isolation, auditability, and confidential-data handling