Portfolio
Selected work
A few projects that best represent how I like to build: high-leverage, technically deep, and grounded in real use.
Most of my work lives at the intersection of systems, tooling, and applied AI. I like projects where the technical depth is real but the outcome is still easy to explain: faster workflows, safer migrations, better products, sharper interfaces.
Outlook Mail Service Modernization
At Microsoft, I've worked on the modernization of Outlook Mail Services, including the Model A to Model B2 transition. The work spans traffic splitting at roughly 350 million requests per hour, OData infrastructure, and decoupling legacy mail flows such as CC.MIME-linked behavior into more maintainable service boundaries. It's the kind of systems work where rollout strategy matters as much as code.
AI-Native Infrastructure
I built internal AI tooling to improve developer leverage: Smart Coverage Analyser reached 62% coverage in six weeks, custom MCP-based workflows reduced friction around Azure DevOps tasks, GRAIL-style impact analysis improved confidence around changes, and OverPowered CLI tied it all together into a faster engineering loop. The broad result was about 5x productivity on the workflows these tools touched.
Dentistry Automation Voice Bot
Built an AI voice automation system for dental clinics using Twilio, Whisper, and GPT-based orchestration. The interesting challenge was not just generation quality, but handling real-world conversational latency, interruptions, and trust. It pushed me toward thinking about AI as a systems problem, not just a model problem.
HireQuotient Data Enrichment
Worked on enrichment pipelines that combined tools like Perplexity and Gemini to improve structured lead intelligence while reducing external API dependence. Smarter orchestration and filtering drove roughly a 75% reduction in API usage without sacrificing output quality.
Video Summarization Research
Co-authored research on video summarization using quaternion transformers, published at IEEE ICPEICES 2024. That project sharpened my interest in the boundary between rigorous ML ideas and practical engineering-the part where research stops being a paper and starts being a system.
Local BizChat
Built Local BizChat for Microsoft's Global Hackathon, an on-device, privacy-first AI assistant for small businesses on Copilot+ PCs. The project used PHI-3 Silica on the NPU and went on to win 3rd place globally. It was one of those rare projects where product framing, systems constraints, and demo storytelling all clicked at once.