02 — Agentic AI in Production · 2025–26
Multi-Agent MCP Server
A multi-agent MCP server that automates BluChip loyalty profile updates across IndiGo's internal databases — cutting turnaround from 12 days to under a minute.
Role
- Design & deployment
Context
- IndiGo
Year
- 2025–26
Stack
- Semantic Kernel, Azure OpenAI, FastAPI, FastMCP
12 days
turnaround before
< 1 min
turnaround after
Prod
deployed at IndiGo
Overview
Updates to BluChip loyalty profiles used to take up to 12 days. I owned the end-to-end design and production deployment of a multi-agent system, exposed through the Model Context Protocol, that handles each request end to end: verifying the customer's government ID with OCR, enforcing PII compliance, and writing the update across multiple internal databases.
Agents are orchestrated with Semantic Kernel on Azure OpenAI, with FastMCP exposing tools behind a FastAPI service. Turnaround dropped from 12 days to under a minute, directly improving operational throughput for IndiGo's loyalty team.
What I built
- 01Multi-agent orchestration with Semantic Kernel on Azure OpenAI
- 02OCR-based government ID verification before any record is touched
- 03PII compliance built into the workflow, not bolted on afterwards
- 04Tools exposed through FastMCP behind FastAPI, updating multiple internal databases
Architecture
Enterprise work — source is private.
01 Request
Profile update
With customer ID document
02 Orchestrator
Planner agent
Semantic Kernel · Azure OpenAI
03 Checks & actions
ID verification
OCR · government ID
PII compliance
Policy guard
Profile update
Write agent
04 Tools
MCP server
FastMCP · FastAPI
05 Systems
Internal databases
BluChip loyalty records