AI Software Engineer and tech lead with 14+ years building and operating large-scale distributed systems serving hundreds of millions of users. Specialized in agentic AI systems — multi-agent orchestration runtimes, LLM tool-use and function calling, Model Context Protocol (MCP) integrations, agent-to-agent (A2A) communication, RAG pipelines, memory systems, and evaluation loops. Strong foundation in event-driven architecture, fault tolerance, idempotency, and horizontal scalability. Proven track record taking AI products from 0→1 concept through GA, and making binding architectural decisions in ambiguous, fast-moving environments. Passionate about shipping reliable, low-latency AI that measurably improves user productivity at scale.

• 🔭 Currently: Principal Software Engineer · AI Tech Lead @ Microsoft AI (Teams Platform) — building agent APIs, SDKs, and the multi-agent orchestration runtime that serves 300M+ daily active users.
• 🌐 Portfolio & deep dive: rajuroopani.github.io

• Agent Extensibility Platform & Orchestration Runtime — Tech lead for Agent APIs, SDKs, and client experiences that let internal and external developers build production-grade agentic systems in Teams. Architected and delivered a real-time distributed multi-agent orchestration runtime supporting A2A, MCP, at-mention, context propagation, turn-taking protocols, and human-in-the-loop intervention — event-driven, with partial-failure isolation, idempotent retries, and horizontal scale-out delivering reliable, low-latency AI interactions to 300M+ Teams DAU.
• Teams Meetings Platform Framework & SDK — Led integration of AI agents into Teams Meetings, providing real-time insights and contextual assistance to millions of daily active users.
• Cross-Functional Technical Leadership — Scope ambiguous work independently, evaluate build-vs-buy trade-offs, and make binding architectural decisions across Engineering, Product, Design, and AI Research without escalation.
• Modern Project Online (project.microsoft.com) — Full-stack architecture and delivery of a cloud-native collaborative platform on React/TypeScript, GraphQL, and Azure for millions of enterprise users; cut page-load time 30% via virtualization, memoization, and batch-update strategies.

• Microsoft AI — Principal Software Engineer · AI Tech Lead, Teams Platform · Jul 2018 – Present
• Envestnet Yodlee — Member of Technical Staff, Platform Engineering · Apr 2016 – Jul 2018
  High-throughput financial aggregation platform at 10M+ transactions/day (Java Spring Boot + Apache Kafka); framework-agnostic integration bridging 100+ heterogeneous bank APIs into a unified runtime; OAuth 2.0 client framework for PSD2-compliant EU Open Banking.
• ARM Embedded Technologies — Software Engineer, Build & Automation Tooling · Jan 2012 – Apr 2016
  Automation tooling for Physical IP modeling & validation — reduced manual verification effort ~60% and established build patterns reused across R&D teams.

🎓 B.E., Electronics & Communications Engineering — Osmania University, Hyderabad, India (2007 – 2011)

Not a logo wall — here's the stack I build with, and the role each piece plays in my work.

Stack	How I apply it
	First-class context & tool interface so agents and 3rd-party tools interoperate without bespoke glue
	Protocol for inter-agent communication and coordination across the multi-agent runtime
	Distributed runtime — turn-taking, context propagation & human-in-the-loop control at 300M+ DAU scale
	Grounds LLM responses in live, domain-specific context for accuracy and trust
	Golden-set regression eval, telemetry-driven quality gates & distributed tracing — the contract that lets models change without product regression
	Foundation models behind a thin, provider-agnostic layer — swap providers, route by task complexity, unify telemetry

Stack	How I apply it
	Decoupled, asynchronous coordination — the backbone for resilience and elastic scale
	Independently deployable services with clear ownership and contained blast radius
	Partial-failure isolation and idempotent retries designed in as defaults, not afterthoughts
	Stateless scale-out patterns proven at hundreds of millions of users
	Developer-facing SDKs, API governance, and onboarding that make the right thing the easy thing

Stack	How I apply it
	Agentic systems, LLM orchestration, eval harnesses, and AI tooling
	High-throughput Teams platform services
	Built event-driven financial-aggregation APIs handling 10M+ transactions/day
	Type-safe full-stack development across UI and services
	Lightweight API/tooling runtime and relational data access

Stack	How I apply it
	Component architecture for cloud-native collaborative platforms (Project Online)
	Optimized the Data Grid rendering pipeline for a 30% page-load reduction
	Typed, efficient client–server data contracts

Stack	How I apply it
	Primary cloud — Service Bus, Functions, Cosmos DB, and Azure OpenAI
	High-throughput event streaming and reliable enterprise messaging
	Containerized, reproducible packaging for every service
	Document and relational stores chosen to fit the access pattern

High availability, fault-tolerance & horizontal scalability — as first principles.