Available for Principal Architect / Tech Lead roles

Lei Ma
Building AI Systems
at Massive Scale

Hands-on System Architect who leads a 40-person engineering team to design and ship systems at massive scale. As the architect of MSN.com, I lead backend services design, partner with Data Scientists to productionize ML models, and personally code key infrastructure — 400K requests/sec, 2 billion users, $1.3B annual revenue, 2PB of data daily.

400K/s
Peak QPS
$1.3B
Annual Revenue
2B
Global Users
2PB/day
Data Processed
// what_sets_me_apart

The Architect Behind MSN.com

As the architect of MSN.com, I lead the engineering team to design backend services, feature store, user profile, and ranking infrastructure end-to-end. I partner closely with Data Scientists to productionize User Understanding, Document Understanding, and ranking models onto the production platform — powering 400K requests per second, 2 billion users, $1.3B in annual revenue, and 2PB of data daily. I also personally developed key features for Microsoft Copilot — Daily Briefing, Discover, podcast integration, and personalized news. On top of that, I hand-coded a C# workflow engine and a multi-agent AI platform on MAF + GPT-5.4. Finding someone who can architect at this breadth, lead teams, partner with DS, and still write production code — from large-scale services to GenAI agents — is extremely rare.

🌐
MSN.com Backend Architect
Lead team on services, feature store, ranking; partner with DS on model productionization
🤖
GenAI + Agent Platform
C# workflow engine, MAF multi-agent system, 100K+ daily AI content
🧠
Large-Scale Services
Ads Platform ($B revenue), MSN.com backend, distributed systems at extreme scale

What I've Built

Every system personally architected, designed, and coded.

🌐

Lead Architect of MSN.com Backend

As the architect of MSN.com, lead the engineering team to design the full backend: serving layer, feature store, user profile, data pipelines (2PB/day), ranking infrastructure, and telemetry. Partner with Data Scientists to productionize User Understanding, Document Understanding, and ranking models onto the production platform. 400K requests/sec peak, 2 billion global users, $1.3B annual revenue.

400K QPS $1.3B REVENUE 2B USERS 2PB/DAY
💬

Microsoft Copilot Features

Personally developed Daily Briefing and Discover for Microsoft Copilot. Brought podcast functionality and personalized news to the Copilot platform, integrating MSN's content intelligence and ranking infrastructure into Microsoft's flagship AI product.

PERSONALLY DEVELOPED
🤖

Multi-Agent AI Platform

Hand-built on Microsoft's AutoGen Framework (MAF) + GPT-5.4. Pluggable Skills system, hierarchical SubAgent orchestration, autonomous multi-step task execution with human-in-the-loop oversight. Driving the next generation of Microsoft's AI experiences.

PERSONALLY CODED
⚙️

C# Workflow Engine

Production-grade C# execution engine inspired by N8N. Data Scientists design workflows visually and the engine runs them at scale with zero hand-written code. 10x faster experiment-to-production velocity, zero scalability bottlenecks.

PERSONALLY CODED
📈

100K+ Daily AI Content

End-to-end multimodal GenAI platform producing 100K+ pieces of content daily. AI-generated content now surpasses traditional curated feeds in impressions, CTR, and revenue — a paradigm shift for Microsoft's content surfaces.

OUTPERFORMS STATIC FEEDS
🔍

Multimodal Search Engine

CLIP-based Image2Text embeddings + ANN indexing + two-stage semantic ranking. Optimized to P99 < 20ms latency, powering real-time visual search across Microsoft's content surfaces.

P99 < 20MS
🔒

Project Florida — Federated Learning for Ads

Architect and engineer of Project Florida, Microsoft Research's cross-device federated learning platform. Designed the click-to-deploy orchestration infrastructure and device SDKs enabling privacy-preserving ML training across millions of devices for Ads targeting — without raw user data ever leaving the device.

ARCHITECT & ENGINEER MS RESEARCH

From Large-Scale Services to GenAI

A 20+ year journey through the entire stack.

Microsoft Principal Architect & Engineering Leader
Jan 2023 – Present  •  Redmond, WA
Architect of MSN.com backend and key contributor to Microsoft Copilot. Leading the 40-person engineering team on system design, partnering with Data Scientists to productionize ML models, personally developing Copilot features, while remaining deeply hands-on in architecture and coding.
  • MSN.com Backend Architecture LEAD ARCHITECT — Lead engineering team to design backend services, feature store, user profile, data pipelines (2PB/day), and ranking infrastructure. Partner with DS to productionize User Understanding, Document Understanding, and ranking models. 400K req/sec, 2B users, $1.3B revenue.
  • Microsoft Copilot Features PERSONALLY DEVELOPED — Developed Daily Briefing, Discover, podcast integration, and personalized news for Microsoft Copilot. Integrated MSN's content intelligence and ranking into Copilot's platform.
  • Multi-Agent AI Platform PERSONALLY CODED — Built on MAF + GPT-5.4 with pluggable Skills, hierarchical SubAgents, and autonomous task orchestration.
  • No-Code Agentic AI Platform PERSONALLY ARCHITECTED — N8N + custom C# workflow engine. Applied Scientists deploy GenAI pipelines with zero production code. Weeks → hours.
  • GenAI Content Pipeline PERSONALLY DESIGNED — 100K+ daily multimodal content. 3× throughput, ~40% cost reduction. Outperforms static feeds in revenue.
  • Multimodal Search Engine PERSONALLY BUILT — CLIP + ANN + semantic ranking. P99 < 20ms.
  • Unified Personalization Platform — Recommendation engine serving billions of users. Unified fragmented architecture.
  • Platform Reliability — 99.99% availability. Embedded Responsible AI safety policies into serving layer.
Microsoft Principal Software Engineer / Technical Lead
Jul 2016 – Jan 2023  •  Bellevue, WA
Lead Architect for Microsoft's Ads Platform backend. Deep E2E knowledge of the entire advertising workflow. Personally re-architected core systems, and partnered with DS to build AI-powered advertising features.
  • Ads Platform Re-architecture PERSONALLY EXECUTED — Single-handedly re-architected Ads backend: 30% performance improvement, 25% database cost reduction. Redesigned query patterns, connection pooling, and caching.
  • Project Florida — Federated Learning ARCHITECT & ENGINEER — Designed and built Microsoft Research's cross-device FL platform: click-to-deploy orchestration infra + device SDKs. Enabled privacy-preserving ML for Ads targeting across millions of devices.
  • AI-Powered Ads Features PARTNERED WITH DS — Auto-generation of ad creatives from product pages, ad performance prediction, account spend forecasting. Personally built the ML model serving platform powering these features.
  • E2E Ads Workflow Expertise — Deep knowledge across the full pipeline: campaign management, ad serving, auction, billing, targeting, reporting. Drove cross-system architectural optimizations.
  • Model-to-Production Integration — Standard patterns for safe ML model deployment. Led cross-functional teams for AI-driven features with GDPR-compliant targeting.
Microsoft Senior Software Engineer
Mar 2012 – Jul 2016  •  Redmond, WA
Core systems engineer on Windows OS. Low-level networking and cross-device protocols shipped to 1B+ devices.
  • Windows 10 Cross-Device Sync Protocol — TCP/UDP hybrid transport optimized for mobile power constraints. Shipped to 1B+ devices globally.
  • Universal Cellular Certification Kit — Automated protocol testing for Qualcomm, MediaTek, Intel. Months → weeks.
Microsoft Engineering Manager
Oct 2005 – Mar 2012  •  Beijing, China
Built high-performing engineering teams delivering core networking and multimedia features across Windows Phone and Embedded OS.
Objectiva Software Solutions Software Developer
2000 – 2005
Early career in enterprise software development. Foundational systems programming and engineering skills.

Hands-On Tech Stack

Not just keywords — technologies I've personally shipped in production.

Systems & Languages
C#/.NET C/C++ Python Systems Programming Networking
AI/ML Infrastructure
GPT-5.4 GPT-4 Phi AutoGen (MAF) RAG Pipelines CLIP/Image2Text ANN Indexing Safety Classifiers
Agent & Workflow Systems
Multi-Agent (MAF) Skills System SubAgents N8N Orchestration Custom Workflow Engine Agentic Chains
Distributed Systems
Microservices High-QPS Serving Real-time Inference Lock-free Structures Cost-aware Routing
Platform & Infrastructure
Azure Kubernetes gRPC Event-driven Observability CI/CD
Leadership & Process
40+ Person Org Technical Strategy Cross-functional Responsible AI Patent Holder

Let's Build Something

Open to Principal Architect, Technical Fellow, and Staff+ Engineering roles.