Vandna Sharma · Bengaluru, India
I build AI systems that make engineering work clearer.
I'm a senior backend and applied AI engineer. My work sits where real-world engineering problems meet practical AI: systems that are useful, observable, and built to earn trust.
What I care about
Useful over flashy.
Clear failure modes.
Technology in service of people.
Featured work
A closer look at my work.
MCP-powered natural-language automation for network infrastructure
An open-source MCP server that helps engineers work with a complex load-balancer API without memorising its schema.
Featured at VMware Explore US 2025.
02Multi-agent AI platform for product diagnostics
An agentic investigation system that turns support bundles, logs, and configuration files into clear root-cause reports.
Investigation time reduced from hours to under 2 minutes.
03Code intelligence with AST and call-graph analysis
A structured code-search layer that helps an RCA system find related code beyond text similarity.
30,000+ symbols indexed across Go and Python.
04GenAI operations assistant for Avi load balancer
A conversational assistant that turns Avi's application analytics and documentation into plain-English answers and guided recommendations.
Featured at VMware Explore US 2025.
05RAG-powered assistant for product documentation
One interface for documentation answers, live system insight, and relevant training material.
Presented at VMware Explore Singapore 2023.
06ML-powered CI/CD log triage
A failure-classification system that identifies recurring Jenkins issues and routes them to the right place.
80% less manual triage for DevOps and SRE teams.
07Agentic software failure investigation for CI/CD pipelines
An agentic system that reasons over test-run logs and automation code together to explain why a test failed, verified by a separate critic agent before it's shown to anyone.
Currently building.
Writing
Thinking in public.
I write on Substack about the less glamorous but essential parts of applied AI: architecture, security, evaluation, and adoption.
Browse the full archive →We Spent Months Building an AI Harness. Then the Model Started Ignoring It.
Read on Substack ↗What your coding assistant really does when you ask about your codebase
Read on Substack ↗The AI read every page. It still answered from the wrong one.
Read on Substack ↗AI Gateways, DLP and What Comes After WAF
Read on Substack ↗What it actually takes to know if your AI agent is working
Read on Substack ↗Beyond the work
When I'm not untangling a system, I'm usually making something: an art project, a dance step, a yoga practice, a travel plan, or a small learning adventure with my kids.
More about me