Hey, I'm Vandna 👋
I spend my days building multi-agent AI systems, RAG pipelines, and applied AI features — and the distributed backend platforms that hold them all together. These days that means shipping production AI at Broadcom (formerly VMware) out of Bengaluru, India, and writing about what I learn along the way.
Years of Engineering Experience
Production AI Systems Shipped
Articles Published
Conference Talks
Ask me anything, or just have a look around
I'm a senior backend and applied AI engineer with 12+ years of experience building production AI systems, multi-agent workflows, RAG pipelines, and distributed backend platforms.
[email protected]My career has moved through Mojo Networks, Arista Networks, and now Broadcom (formerly VMware), and along the way I've grown from building backend REST APIs to designing agentic AI platforms that reason over real production data. I care most about the point where solid backend engineering meets genuinely useful applied AI — systems that are fast, reliable, and actually save people time.
At Broadcom, I work on an agentic root-cause-analysis platform built with Google ADK, Gemini 2.0 Flash, and Qdrant that analyzes customer support bundles and generates root cause reports, cutting investigation time from hours to under 2 minutes. I also built a code intelligence layer using tree-sitter AST parsing that indexes 30,000+ symbols, and an open-source MCP server (built with FastMCP) that lets engineers interact with network infrastructure in natural language — featured at VMware Explore US 2025. Earlier work includes a RAG-based AI assistant for product documentation, presented at VMware Explore Singapore 2023, and an ML-powered CI/CD log analyzer that reduced manual triage by 80%.
Before that, at Arista Networks I built analytics dashboards and led PostgreSQL/MongoDB cluster migrations, and at Mojo Networks I built backend REST APIs for enterprise wireless network management using Python, Django, and PostgreSQL.
I hold an M.Tech in Computer Science (Data Engineering) from IIIT Delhi (2011–2013), and I'm based in Bengaluru, India.
Dec 2019 – Present
Aug 2018 – Dec 2019
May 2013 – Jul 2018
M.Tech, Computer Science (Data Engineering) — IIIT Delhi, 2011–2013.
The languages, frameworks, and infrastructure I reach for most often when building production AI systems and backend platforms.
A selection of the AI systems and backend platforms I've built over the years.
An agentic platform that analyzes customer support bundles and generates root cause reports automatically. It cut investigation time from hours down to under 2 minutes, secured behind OKTA SSO and observed end-to-end with LangFuse.
A code intelligence layer that indexes 30,000+ symbols using tree-sitter AST parsing. Search runs through a 4-stage hybrid pipeline — exact match, LLM query expansion, semantic search, and call graph traversal — to find the right code fast.
An open-source MCP server that lets engineers interact with Avi load balancer's REST API in natural language, featured at VMware Explore US 2025. It uses RAG-based self-correcting error recovery to keep automated workflows resilient.
A multi-source RAG system for product documentation, presented at VMware Explore Singapore 2023. It combines docs Q&A with live API function calling for controller introspection and training video recommendations.
A machine learning model that predicts Jenkins test-failure categories directly from log patterns, cutting manual triage time by 80%. Confident matches are auto-assigned to Jira tickets, closing the loop end to end.
Notes on AI security, agentic systems, and applied AI engineering. I publish on Substack.
See all posts on Substack
vandnasharma1.substack.com
How model upgrades force a rethink of what an effective AI control mechanism even is.
Read on Substack →The structured memory layer that persists across conversations.
Read on Substack →Security approaches focused on behavioral intent rather than syntax-based rules.
Read on Substack →Prompt injection, jailbreaks, and internal threats to AI systems.
Read on Substack →Why traditional deterministic security models need to evolve for AI contexts.
Read on Substack →Evaluation metrics and their limitations for assessing AI agent performance.
Read on Substack →How different questions follow distinct processing paths within coding assistants.
Read on Substack →Tracing token processing and attention mechanisms from question to answer.
Read on Substack →Comparing vector-based and vectorless RAG retrieval approaches.
Read on Substack →Production challenges beyond prototype development for enterprise AI systems.
Read on Substack →Five critical patterns to watch for with AI-assisted coding tools.
Read on Substack →The underlying workflows powering tools like Cursor, Claude Code, and Devin.
Read on Substack →Read more on Substack
vandnasharma1.substack.com
Happy to talk about applied AI, backend systems, or anything in between. The easiest ways to reach me are below.