About Me

Career thread below

Jump to experience

Microsoft

Production AI agents and the Azure services they run on

growIndigo

ML pipeline that took crop-classification accuracy from 80% to 93%

IIIT Delhi

CS and Applied Math (Honors), concurrency research

Dhruv Mishra - Software Engineer at Microsoft

Hey, I'm Dhruv 👋

I'm a Software Engineer at Microsoft. I build production AI agents and the Azure backend services they run on.

Most of my recent work is LLM incident-triage agents. MCP servers give the agent Kusto, service-health, and config tools, so on-call engineers spend less time on routine incidents. I also cut $240K a year from the Azure bill of a backend that handles over 7 billion requests a day.

I mostly write C#, TypeScript, and Python, with C++ and Kotlin when the job calls for them. On my own time I build voice agents: Invoke hands coding tasks to Copilot CLI agents when you ask out loud, and Jarvis negotiates freight rates over a phone call.

I maintain Fluent UI Android, Microsoft's open-source Compose library. Outside code, I'm into strength training, PC overclocking, chess, and reading about AI systems and longevity research.

I graduated with Honors in CSAM from IIIT Delhi, and spend time honing my skills through competitive programming.

💬 Reach out: [email protected] • +91-9599377944

📄 For more details, check out my resume.

Curious about my work? Ask me anything.

Click here to chat with me

Pinned career thread

Experience Timeline

Building production AI agents and the Azure backend services they run on.
Incident-triage agents that cut manual on-call triage by 40%, and $240K/year in Azure savings on a backend serving 7B+ requests a day.
- Built and shipped LLM incident-triage agents for on-call engineers. MCP servers give the agent Kusto, service-health, and config tools, with fallback logic and human escalation. Manual low/mid-severity triage dropped 40%, measured by replaying resolved incidents.
- Cut Azure spend by $240K a year (15%) on a 7B+ requests/day Service Fabric backend by raising the Redis cache hit ratio from 31% to 80%, tuning Cosmos DB and log retention, and retuning autoscaling across 12+ VM Scale Sets, all within the 99.99% availability target.
- Shipped PDF encryption for Word, Excel, and PowerPoint in Office for the web by sandboxing a C++ encryption SDK and routing its Azure RMS calls through a C# proxy over IPC.
- Cut Office Android cold-start and file-open latency by about 500ms with custom baseline profiles, and Excel's loading shimmer from 300ms to 3ms.
- Primary maintainer of Fluent UI Android, Microsoft's open-source Jetpack Compose library: 10 releases used by Copilot, Office, Outlook, OneDrive, and Teams, plus its security compliance.
C#TypeScriptPythonC++KotlinMCPKustoService FabricRedisCosmos DBVM Scale SetsJetpack Compose