ROLE SUMMARY

Leads the design of multi-agent systems across products. Owns MCP and A2A integration standards, agent orchestration and state management patterns, and production guardrails. Primary technical authority for agentic system architecture for the Product.

KEY RESPONSIBILITIES
Architect multi-agent workflows, semantic routers, and autonomous systems using LangGraph, AutoGen, or CrewAI
Design and govern MCP and A2A frameworks for secure agent-to-tool and agent-to-agent communication
Implement orchestration layers managing agent memory, context, planning, and task execution
Define guardrails, safety policies, and hallucination-prevention protocols for production agentic AI
Design observability pipelines to trace, evaluate, and monitor multi-agent decision chains at scale
Mentor Senior Agentic AI Engineers and MCP Integration Engineers

TECHNICAL SKILLS
Languages & Frameworks: Python, TypeScript, Java; LangGraph, AutoGen, CrewAI, LlamaIndex
AI Coding Tools: Claude Code, GitHub Copilot, OpenAI Codex, Cursor
Protocols: MCP,



A2A connector architecture and agent card design
LLMs & GenAI: RAG, hybrid retrieval, prompt engineering, vector databases (Pinecone, pgvector, Weaviate, Qdrant)
Structured Outputs & Function Calling: JSON schema enforcement, tool schema authoring
Prompt Caching: Anthropic / OpenAI caching for latency and cost optimisation
Sandbox Execution: E2B, Modal — protected agent code execution setting design
AI Security: Guardrails AI, NeMo Guardrails, prompt injection prevention, access controls
Observability: LangSmith, Arize Phoenix; agent harness testing, evals (Ragas, DeepEval, HELM)
Cloud AI: AWS Bedrock, Azure AI Foundry, GCP Vertex AI
HITL Design: human escalation paths and graceful agent handoff patterns
Agent Orchestration: Task Decomposition, Subagent Delegation, Progressive Context Loading
Interactive Planning & Grill-Me Prompting: constraint validation before any code or config is committed

NICE TO

📌 Agentic Systems Architect Bengaluru (India)
🏢 Cognizant
📍 India