Building multi-agent systems, intelligent workflows and enterprise AI systems with LangGraph, LangChain, MCP, Python and Azure.
Agentic AI becomes useful when models are connected to planning, tools, state, specialized agents, validation and controlled execution.
Generation is the starting point.
Decide, act and observe.
Connect intelligence to action.
Carry workflow context forward.
Coordinate specialized capabilities.
Evaluate, observe, secure and recover.
Choose a goal and watch the observable execution cycle: plan → tool → observe → validate → replan. No private model reasoning is exposed.
Run the loop to see observable system events. This portfolio intentionally does not expose hidden chain-of-thought.
I’m Sayeri Biswas, an Agentic AI Engineer and Technology Analyst at Infosys with nearly five years of enterprise engineering experience.
I build LLM-powered applications, multi-agent systems and AI-driven workflow automation, with a focus on orchestration, state management, planning and replanning, tool integration, evaluation, observability and production engineering.
Enterprise EDI AI Assistant supporting EDI SDLC workflows and enterprise knowledge retrieval through multi-agent orchestration, MCP-based tool integration and workflow execution.
LLM-based classification, extraction, summarization and workflow automation reduced manual effort by 76% in the documented enterprise workflow.
The Digital Brain brings orchestration, specialized capabilities, enterprise knowledge retrieval, tool integration, validation and human approval into one observable workflow.
Healthcare EDI capabilities represented in the documented engineering stack.
Choose a scenario and observe the system move through planning, tool use, state updates and validation — without exposing private model reasoning.
Orchestration, state, planning, execution and human control form the engineering layer around the model.
Sequential and parallel workflows, conditional routing, specialized agents and controlled multi-step execution.
Shared context and persistent workflow state allow execution to continue across multiple steps.
Planning and replanning coordinate complex tasks and adapt workflow execution.
Approval points provide controlled intervention where the workflow requires human oversight.
Connect agentic workflows to enterprise APIs, databases, data sources and tools through MCP-based integration.
Custom MCP tools, discovery, integration and permissions/authorization become part of the agent's observable execution path.
Enterprise RAG pipelines support contextual grounding through retrieval, filtering, reranking and evaluation.
Evaluation, observability, guardrails and recovery are part of the engineering surface.
Security controls can be visualized as a trust boundary around untrusted inputs, agent authorization and tool execution.
From LLM-based classification, extraction, summarization and workflow automation.
Click a capability group to inspect the technologies behind the systems.
RCC Institute of Information Technology
Infosys Limited
Professional credential in Agentic AI.
Professional credential in Python.
Outstanding project contributions.
Interested in Agentic AI, intelligent systems, enterprise automation, or AI engineering?