Principal Software Engineer · Microsoft Fabric
I'm the tech lead and architect for capacity management in Microsoft Fabric. My work spans distributed systems, multi-tenant platforms, enterprise security, and the infrastructure behind AI agents.
Outside my day-to-day platform work, I build open-source developer tools, contribute upstream, speak about reliability and agent workflows, and take part in peer review and judging.
Platforms, infrastructure, and applied machine learning.
– Present
At Microsoft Fabric, I lead capacity management: how a shared cloud platform allocates compute, measures consumption, and bills for it. My work connects that control plane with enterprise security and service reliability, and extends into MCP-based tools for engineering and incident investigation.
–
I built conversational systems for Amazon Customer Service, connecting pluggable ML models to a TensorFlow serving layer. We piloted the platform with support agents worldwide.
–
At AWS, I worked with hardware and infrastructure teams on how data-center racks were allocated. Turning those planning constraints into software reduced manual work and made better use of the hardware.
Ira A. Fulton Schools of Engineering
Indian Institute of Information Technology, Design and Manufacturing, Kancheepuram
Selected open-source work.
Featured project
A managed local runtime for the tools behind AI agents. MCP Pacemaker separates tool-server lifecycles from the editors and CLIs that use them, bringing supervision, authentication, configuration, and diagnostics into one place.
I built it to make an agent workspace easier to operate across sessions and hosts, instead of treating every tool server as a disposable editor subprocess.
A Markdown and Mermaid reader for VS Code and the browser. One shared renderer brings readable diagrams, syntax highlighting, reading themes, and a live table of contents to both environments, with all rendering kept local.
Control macOS system idle sleep from the menu bar, the command line, or a local web interface.
Protocol work, community participation, and implementation contributions.
Participate in the Skills Over MCP community and the SDK, Transports, Interceptors, and Agents channels. Contributed C# SDK implementations for the Skills extension and Native AOT array handling; both submissions are under review.
Participate in the A2A community and SDK discussions. My merged Java SDK contribution preserves explicit JSON nulls in gRPC mappings.
Java SDK #1174
Participate in the GenAI instrumentation and semantic-conventions communities, with contributions under review in GenAI semantic conventions and Python instrumentation.
GenAI semantic conventions #540 #541 #542
Python instrumentation #793
Participate in the Core working-group community, including SLIM and Agent Directory channels. My SLIM receive-recovery contribution is merged, with additional OASF and CSIT work under review.
Contribute to agent-protocol and charter discussions on dialog lifecycles, uncertain execution outcomes, authorization context, and bounds on agent actions and spending. Also follow the WIMSE and DAWN mailing lists.
Contributed merged MCP server lifecycle fixes to VS Code and improved the packaging of Copilot CLI runtime assets in the GitHub Copilot SDK.
Copilot SDK #2557
Reliability and agent infrastructure.
Separating alert routing from investigation, and keeping human judgment where it matters.
Feature flags and guardrails for agentic AI in engineering workflows.
Peer review and judging.
Continual Learning for Enterprise AI Agents.
Judge for the Innovation, Leadership, and Business programs. Received the Innovation judge credential in September 2026.
Find me on LinkedIn for professional conversations, and across both GitHub accounts for projects and upstream contributions.