Yang Qiu

Selected work

1. Haiclaud · 2025 – now

   MoiAI: an agent that runs on your machine

   A desktop AI agent for professionals, local and private, on macOS, Windows and the web. Execution loop, skill framework, inference scheduling and telemetry — all built from first principles.

2. Microsoft · AI Platform

   The groundwork of large models

   Helped build the infrastructure behind OpenAI and Microsoft models: a three-tier checkpoint store that keeps hyperscale training continuous (Nebula), and global rate limiting and quotas that keep inference stable (MaaS).

3. Microsoft · M365 Copilot

   The runtime of Copilot

   Owned the runtime architecture of the Microsoft 365 Copilot Platform — orchestration, LLM integration and compliance — bringing the assistant into the daily flow of Word, Excel, Teams and Outlook.

4. Huawei · 2012 Labs

   HarmonyOS IDE and Personal DevCloud

   Shaped the engineering architecture of the HarmonyOS IDE through its on-schedule bring-up, and led a developer cloud that cut a new engineer’s path to first commit from days to ten minutes.

5. Baidu

   Data and ads for the mobile era

   Designed the storage architecture of Baidu’s mobile data platform and led three generations of the Fengchao mobile ads system through the rapid rise of the mobile internet.

Experience

1. Haiclaud Technology Limited (Hong Kong)

   Nov 2025 – Present

   Co-Founder / CTO

   Leads the technical architecture and 0-to-1 delivery of MoiAI, and owns enterprise AI-native transformation — scaling LLM and agent capabilities across R&D, design, e-commerce, content production and supply chain.

   MoiAI — desktop AI agent (product & technical architecture)

   A desktop AI agent for professionals with local, private execution across H5, macOS and Windows (live.moi-ai.com).

   • Benchmarked against Claude’s Skill framework and implemented its complete skill capabilities, building a continuously evolving agent ecosystem.
   • Re-architected the agent pipeline (planning → tool invocation → execution → feedback), improving response speed and multi-step task performance.
   • Designed inference scheduling for high-concurrency, high-TPM workloads.
   • Reduced labor cost through automated task orchestration and multi-agent collaboration.
   • Owned telemetry end to end, closing the collect–analyze–iterate loop.

   Enterprise AI-native transformation

   • Leads enterprise-wide transformation centered on LLMs and agents across the full value chain.
   • Built the “Vibe Coding” AI engineering team from 0 — an “LLM + AI pair-programming” process that compresses delivery from weeks to days.
   • Brought frontier models to design, e-commerce and short-video channels (Douyin, Kuaishou) for image and video generation at scale.
   • Built an intelligent supply-chain system for demand forecasting, production planning and scheduling.
   • Established unified AI application and data governance with reusable enterprise AI assets.

2. Microsoft (Asia Engineering, Beijing)

   Jul 2019 – Nov 2025

   Business Architect / Project Lead (Level 64)

   Worked on top-tier AI infrastructure and the Copilot platform, directly serving LLM training and inference for OpenAI and Microsoft.

   M365 Copilot Platform — Business Architect (Runtime)

   • Owned the runtime core architecture — orchestrator, LLM API integration and data compliance.
   • Led integration and rollout of Copilot across Word, Excel, PowerPoint, Teams and Outlook.

   AI Platform — Project Lead / Architect (training & inference)

   • Nebula: designed a three-tier checkpoint storage architecture for multi-node LLM training — ~50× faster checkpoint saving and minimal-cost resume after crashes, for OpenAI and Microsoft.
   • MaaS: as China Tech Lead, led the runtime engine, inference-framework benchmarking and multi-model loading; designed the Rate Limit Service integration for global throttling and quota protection of OpenAI / Azure OpenAI backends.
   • Parallel Computing: led the framework behind Batch Inference and Many-Models Training; owned org-wide telemetry and data-driven planning.

   Microsoft Ads — China Project Lead / Architect

   • Clarity × UET Tag: led design, review, architecture and monitoring, adding 1M+ users to Microsoft Clarity.
   • Containerization: moved an ads platform serving 100+ countries to Azure Kubernetes (.NET Framework → .NET Core), ~400 physical servers → ~400 elastic containers with self-healing, second-level scaling and ~2× faster diagnosis.

3. Huawei (2012 Laboratories)

   Dec 2016 – Jul 2019

   Module Lead / Chief Architect (Level 18)

   AwardsCorporate-level Quality StarIndividual Project Breakthrough Award

   HarmonyOS IDE (team of 6)

   • Owned research, design and architecture of the project-management module; enabled single-project multi-platform builds for the on-schedule system bring-up.

   UK security-certification build tree (distributed backend)

   • Led the distributed backend (service, cache); container scaling cut reverse-parsing time from hours to minutes.

   Personal DevCloud (team of 25)

   • As project lead and chief architect across router and Mate / P-series teams, cut time-to-first-commit from 7–10 days to 10 minutes — ~6× R&D efficiency.

4. Baidu

   Jun 2012 – Dec 2016

   R&D Engineer → Tech Lead / Architect (T6+)

   Awards2015 Outstanding Talent Award (signed by CEO Robin Li)2015 Best Team Award2016 Special Talent Award

   Mobile Operations Data Platform (team of 8)

   • Built channel-level KPI reporting over PB-scale data and the storage architecture onboarding all Baidu mobile products; anti-fraud mining saved hundreds of millions of RMB.

   Fengchao Mobile Ads (team of 30)

   • Led 3 major architecture evolutions; built ad anti-fraud and channel-rating systems; the app-distribution ads business reached ~¥100M annual revenue.

   Ciyuanfan anime platform (team of 10)

   • Built frontend / backend / strategy teams from 0 and led product, architecture and resources (later merged into Baidu Baijiahao).

5. UC Browser (now Alibaba UC)

   Mar 2011 – Jun 2012

   Senior R&D Engineer

   Box Computing search

   • Designed and built the “Box Computing” search directory, speeding up data push and expanding recall.

6. SouLv.com

   Jul 2008 – Mar 2011

   Java Engineer / Senior Java Engineer

   Ads, analytics and logging

   • Owned site-wide ad serving, analytics and logging; built a real-time log-collection service and data system from 0.

Education

B.Eng. in Communication Engineering

Keywords

LLM training / inference / fine-tuningAI agentsCopilot platformDistributed systemsCloud native / KubernetesHigh concurrency / high TPMBig dataAd-serving architectureTelemetryDevOps

Contact

Email: qqyhdt@gmail.com · GitHub: github.com/qyhdt