Hi, I'm Niketan.
First things first, my resume is publicly available here.
I love building automation that makes developers' lives easier. I'm happiest when I'm solving real workflow problems at scale - the kind where manual processes are eating up hours and people are drowning in notifications they don't need.
I like to own my systems end-to-end: from the Azure infrastructure and OAuth flows to the Python backend and the notifications that land in Teams. I have no qualms about diving into legacy code or rewiring monitoring systems that cry wolf. In my experience with developer productivity tools, the biggest wins come from understanding what actually slows teams down, then building pragmatic automation around it.
I'm very comfortable with Python, Django, Azure (Cosmos DB, OpenAI, Active Directory), OAuth 2.0, and building AI-powered automation using LLMs and RAG. I've also written production C# at Microsoft, but it's not my comfort zone (yet).
Right now, I'm part of Microsoft's Developer Productivity team, helping thousands of developers ship features faster.
I spend most of my time writing C#, Python, and Kusto and tinkering with AI, building Developer Agents. One of the developer agent I worked on was recently featured in Microsoft Build 2025.
You'll often find me tinkering with event-driven architectures, Azure pipelines, or machine learning recommendation models.
If you want to talk DevEx, system design, AI for developer workflows, Cricket or just about food and travel, I'm always up for a good chat.
I architected the Microsoft Developer Agent, a conversational AI system that lives in Teams and automates the entire pull request lifecycle. It's powered by GPT-4 and Azure OpenAI, and it handles everything from proactive notifications to intelligent code reviews based on historical reviewer feedback patterns. The system was featured at Microsoft Build 2025 for breakthrough innovation in developer productivity.
I built a comprehensive PR velocity ecosystem that serves 2,000+ developers across 250+ global reviewer groups, processing 300+ daily pull requests. The system has an event-driven C# backend with Cosmos DB and a Semantic Kernel AI assistant that actually understands context.
I engineered a traditional ML-based reviewer assignment system with load balancing across geo-distributed teams, and scaled what started as a hackathon prototype into production - cutting PR turnaround by 12% (from 198 to 173 hours).
I also implemented a proactive notification system through a Teams Bot that delivers contextual alerts: build failures, merge-ready status, release tracking, author-waiting notifications, plus batch breach alerts to keep reviewer engagement high.
I developed a dual-dashboard analytics platform. The review dashboard helps reviewers track pending first reviews versus approval-ready PRs. The metrics dashboard shows team performance percentiles, breach tracking against a 36-hour SLA, and first response time analytics.
I also designed a Teams Nudge system that lets PR authors tag reviewer group champions through automated Teams posts with @mentions, integrated with a DRAFT-first workflow to ensure only review-ready code gets submitted.
On the infrastructure side, I redesigned the monitoring architecture for MS Teams by implementing aggregate-level metrics and a custom alerting platform. This eliminated 70% of false alarms - dropping them from 85% to 15% - and improved incident detection time.
I designed a Bluetooth address management platform that solved device interference problems in testing environments by preventing address conflicts between nearby devices. The system served the entire Bluetooth division at Qualcomm. I was awarded Qualstar recognition for eliminating manual address blocking processes and delivering faster testing workflows.
I built and scaled test optimization solutions through smart grouping and intelligent clustering algorithms, cutting redundant runtime by around 34%. I also led a cross-functional vendor team to scale a 3G device log parser for expanded scenarios, improving regression detection accuracy from 63% to 85% through automated checkpoint implementation.
I built an automated 3G device log parser for regression identification, reducing manual analysis by 34%.
Electronics and Communication Engineering (CGPA: 8.92/10.0)