Open to Staff and Principal GenAI engineering roles
The hard part of AI isn't the model. It's everything around it.
I'm Vinay, a Staff Software Engineer at Intuit with 14 years across five companies. I built the agentic voice assistant for a support line handling 6–7M calls a year, and the evaluation harnesses, retrieval and guardrails that make generative AI trustworthy in production.
IntuitMicrosoftDocusignOracleSAP
14yrs
shipping distributed systems and AI
4M+
GenAI datapoints evaluated a year at Intuit
30+
interview loops run at Intuit, mostly system design
1
granted patent: data privacy as a cloud service
Selected work
All case studies →Intuit · Staff SWE · 2026
Voice assistant for customer support
Replaced a rule-engine IVR handling 6–7M calls a year with an agentic system on LLMs and RAG over 10,000+ live articles. Voice platform chosen after evaluating three vendors on 20+ parameters. Grounded RAG with hybrid search and reranking, chunking chosen by experiment, latency engineered across the voice path, and evaluation at every layer, speech included, backed by an in-house call simulator and evaluator built in weeks. Routes what it cannot resolve to one of 200+ expert queues. Concept to production in under three months.
Transfer rate ~28% → under 20%Intuit · Staff SWE · 2026
GenAI regression evaluation
A RAG stack shipped changes with no way to see if retrieval regressed. I built a golden dataset from real customer utterances, daily automated runs, embedding-model evals (BGE-M3, Stella) and per-query MRR / mAP / hit-rate explainability.
4M+ datapoints / yr, run dailyMicrosoft · Senior SWE · 2024–25
Reinventing Power Pages with agents
Part of the team moving a flagship product from drag-and-drop to natural language: multiple agents that create, modify and manage web apps, with security and data sovereignty built in for enterprise and government customers.
Multi-agent · enterprise and sovereign-gradeModels are commodities. Judgment is not.
The advantage was never the LLM you picked. It's the eval harness, the golden datasets, and the guardrails that tell you — reliably, daily — when the system is wrong.
How I think
All 38 notes →The Handbook
GenAI engineering, from RAG to production agents
Short lessons on retrieval, evaluation, agents and system design trade-offs, each ending with the line I'd say in a design review.
Open the handbook →What is an AI harness, and why does it matter more than your model choice
For a while, I thought picking an AI product came down to one question: which model is underneath? GPT, Claude, Gemini. Smarter model, better product. It seeme…
RAG & RetrievalRAGA Framework, Part 1: Why and how can you evaluate your LLM App?
Part 1 of a series on evaluating LLM apps with Ragas.
Agents & Frameworks10 AI Agent Design Patterns Every AI Engineer Should Know
LangGraph. CrewAI. AutoGen. OpenAI Agents SDK.