300K+
VMs powered by Azure Local guest management infrastructure
Hi, I’m Dhaval Popat.
Software Engineer @ Microsoft · San Jose, CA
I build cloud systems that scale and AI systems that turn complex problems into reliable software.
I work across distributed systems, cloud infrastructure, virtualization, reliability, and applied AI. Today I build Azure Local at Microsoft. Previously, I worked on EC2 host provisioning systems at AWS.
Selected work
Systems, platforms, and engineering leverage.
A curated view of the work that best represents my current focus. Select a project card for architecture context, scale, and a little more of the story.
3,400+
GPU VMs supported by reusable virtualization architecture
$100M
EC2 CAPEX recovered through fleet wide automated remediation
Platform architecture · Microsoft
Azure Local guest management
Architected and launched a security first, low privilege event driven microservice over Hyper V sockets, processing ~1.4M daily validations at <1 ms latency across 300K+ VMs and 20K+ clusters.
Distributed Systems
Hyper V
Low Latency
Security
View details
AI systems · Microsoft
RAG powered CI/CD platform
Established validation across 13 configurations and ~950 nightly tests; a RAG powered agent automated RCA, bug filing, and draft PR generation, reducing regression resolution from ~8 hours to 1 to 2 hours.
RAG
LLMs
CI/CD
Automation
View details
Virtualization · Microsoft
0 to 1 GPU virtualization
Led design and delivery as SME for 7 teams. The reusable architecture supports 3,400+ GPU VMs across AVD, AI inference, and confidential computing; Hybrid AKS adoption saved 3+ months.
GPU
Virtualization
High Availability
View details
Fleet scale systems · AWS
EC2 capacity recovery and reconciliation
Engineered fleet wide remediation that recovered $100M in CAPEX and architected an IP state reconciliation framework processing 5M+ requests/hour that saved $20M.
EC2
Reconciliation
5M+ req/hr
View details
Applied ML · NYU Langone Health
Computer vision for remotely supervised tDCS
Fine tuned Faster R CNN (Inception v2) in TensorFlow to automate webcam based verification of cranial placement of a tDCS headset for an MS clinical trial enrolling 120 patients.
TensorFlow
Computer Vision
Healthcare
View details
Earlier work
Internships, research, and academic projects
My earlier work spans computer vision, full stack systems, cloud applications, data science, and research. I keep that history available without crowding the main portfolio.
Experience
Where I’ve built.
Microsoft
Software Engineer II
- Led 0 to 1 GPU virtualization architecture as SME for 7 teams, supporting 3,400+ GPU VMs.
- Built applied AI systems, including a RAG powered CI agent and an LLM PR validator using multi sample voting and structural verification with a 93% developer fix rate.
- Delivered zero downtime VM memory resizing, adopted by 700+ enterprise customers and eliminating 1,000+ maintenance windows each month.
- Improved platform reliability with Kusto/ADX self service diagnostics and remediation, reducing MTTR by 45% across 26 engineers.
- Scaled AI assisted engineering by defining reusable MCP tooling, Copilot skills, and agent guardrails; drove adoption through org wide learning sessions and mentored 6+ engineers.
Amazon Web Services
Software Development Engineer
- Recovered $100M in EC2 CAPEX through fleet wide remediation for unsellable host capacity.
- Saved $20M with an IP state reconciliation framework processing 5M+ requests/hour and automatically repairing mismatches.
- Designed asynchronous provisioning workflows with SQS to eliminate repeated manual monitoring and provide end to end lifecycle visibility.
- Expanded EC2 into 4 new AWS Regions by bootstrapping host provisioning microservices responsible for building and software validating hosts.
- Cut production investigation time by ~50% by modernizing real time logging for a Tier 1, high TPS EC2 host inventory service.
NYU Langone Health
Student Research Intern
- Fine tuned Faster R CNN (Inception v2) in TensorFlow to automate webcam based verification of cranial placement of a tDCS headset for an MS clinical trial enrolling 120 patients.
Technical toolkit
What I work with.
Languages
Go · Python · Java · C# · JavaScript · SQL · PowerShell
Distributed Systems
Microservices · Low Latency · High Throughput · Caching · Consistency Models · High Availability · Fault Tolerance
Cloud and Platform
Azure · AWS · Kubernetes · Docker · Hyper V · WMI · ARM / REST APIs · gRPC · CI/CD
AI and Agents
Agentic AI · MCP · RAG · LLM Evaluation · Agent Guardrails · Copilot · Autonomous CI Agents
Data and Observability
DynamoDB · RDS · PostgreSQL · MySQL · SQL Server · Azure Data Explorer · Telemetry & Diagnostics
Current focus
Cloud infrastructure · Virtualization · Reliability · AI assisted engineering · Applied AI
Education and research
Foundations.
New York University
Master of Science in Computer Science · GPA 3.78/4.0
2019
University of Mumbai
Bachelor of Engineering in Information Technology
2017
Publication
Music Playlist Generation Using Facial Expression Analysis and Task Extraction · Springer LNNS
↗
Connect
Interested in cloud infrastructure, distributed systems, or AI developer tooling?
I’m always happy to connect with engineers and teams working on hard systems problems.