<Selected Work/>
Engineering work across GitHub Copilot evaluation, AI-assisted cloud development, Visual Studio, and large-scale services.
# Projects & Experience
GitHub Copilot Model Evaluation
Built cross-platform grading and run orchestration for an LLM evaluation platform supporting 300K+ submitted runs and 10M+ task evaluations per month. Work also includes a self-service grader registry and production run analysis.
Real-World Windows Agent Benchmark
Developed a Harbor-native benchmark of 58 real-world Windows tasks for VS Code Copilot, Copilot CLI, Copilot App, and other company products.
Copilot for Azure Compute and Dev Center
Built an Azure OpenAI and Cosmos DB service that translated natural-language requests into validated, user-approved execution plans for cloud workstations. Delivered the initial VS Code and Visual Studio experience with a two-engineer team in one quarter.
GitHub Copilot for Visual Studio
Co-developed semantic code context using indexed symbols and token-efficient code skeletons. In a 91-case offline comparison, 44.0% of responses met both “no hallucination” and “problem solved” criteria, compared with 31.9% without the context prototype.
Amazon Learning Products
Built AWS infrastructure and TypeScript authorization for Amazon Learn users, then shipped a React frontend and AWS pipeline that reduced third-party learning-software costs by over $4 million annually (30%).
Windows Packaging Service
Extended Windows Packaging and Servicing APIs for a geo-redundant update pipeline serving 1.5 million+ clients every two weeks. A .NET certificate-rotation service reduced legacy service downtime from two hours to under 30 seconds.