Senior Software Engineer · Intuit
I architect the systems between a seller and a closed deal: quote platforms, pricing programs, and the models and agents that recommend the next configuration.
Before Intuit the work was marketplaces, payments, and computer vision at production scale. The current chapter is making quotes win: a platform that stays up, a model trained on closed deals, and agents that can call both.
Architected, trained, and deployed CPQ’s initial production machine-learning system. It serves real-time recommendations through MXS in four modes: best quote, multi-option, industry, and constrained. The model was trained on 12,000+ historical won quotes.
Engineered an MCP server from scratch with FastMCP: 21 tools, covering quote GraphQL plus 18 MXS recommendation tools, with marketplace governance. Claude Code and Cursor agents can query quotes, run the models, compare pricing strategies, and validate inputs against production CPQ. Commerce agents moved to enterprise-wide production GA.
Wrapped that server as the CPQ Quote Plugin, an Intuit Desktop developer extension. Quote lookup by id, contract, or opportunity; ML recommendations; multi-strategy price comparison; catalog browsing; price estimation; and pre-submission validation. It holds official org-wide status on the AI Plugins marketplace, so a team installs it once.
Led the programs that move money: new business, wholesale, amendments, renewals, and deal-lever pricing. Incremental-revenue coverage lands on every IES flow, with a 5,000+ historical backfill so Finance and Sales see live spend and incremental ACV.
Re-architected Quote-Service to 99.99% active-active, built Quote-Async, and migrated the Salesforce data lake onto Quote Clean Entity — from 11,000+ records to 17,000+ on V2.
Columbus Foundation is the write path for seller and product data across Walmart marketplaces: sub-100ms at 25,000+ transactions per minute. Columbus Read serves that data to 100+ consumer teams at 50,000+ transactions per minute, still under 100ms. Alongside it, adaptive MFA drops a second factor to one when the customer’s own interaction pattern says the extra step is friction, not safety.
End-to-end mobile and DTH recharge: AngularJS on the front, Node and Express in the middle, Spring Boot for the business logic, Kafka between them, MySQL underneath. Validation, failure handling, and notifications were part of the transaction, not a follow-up ticket.
Real-time internal monitoring and customer communication, with intelligent routing and multi-channel notifications. React, Spring Boot, Oracle, Azure.
Enterprise video: scheduling, instant meetings, and real-time audio and video. React, WebRTC, WebSocket, Spring Boot, OAuth2, Kafka, Docker, Kubernetes, AWS.
Computer-vision analytics and a central alarm desk. 50+ ML, deep-learning, and AI features for 2,000+ clients. Java, Spring Boot, Node, Angular, React, Kafka, MongoDB, AWS.
The same thread runs through every role: high-volume systems, a team that can ship them, and a habit of replacing a slow path with one people will trust.
The Intuit systems stay inside the company. These are the public experiments: multi-agent research, retrieval, and trading simulations, with code and live demos.
Four agents — Warren, George, Ray, and Cathie — each run a portfolio with a different philosophy, live market data, and a monitoring desk.
A CrewAI crew that researches a company, argues the market, and writes the investment note, with a trace you can replay.
An OpenAI Agents SDK workflow that plans a research task, searches, and writes a long report, then can send it by email.
PDF in, grounded answers out. ChromaDB holds the embeddings; an agent writes the reply. Running on Hugging Face Spaces.
CrewAI agents take opposing sides, research the claim, and run a structured debate instead of a single confident paragraph.
A chatbot grounded in this profile, with OpenAI and Gemini, lead capture, and a notification when someone serious writes.
Specialist agents for technicals, sentiment, and risk vote on a trade. A dashboard watches the portfolio.
Multi-step reasoning, tool calling, RAG, memory, and guardrails — ReAct, chain-of-thought, and prompt design used in production, not as a slide.
Low-latency services, event-driven platforms, and cloud-native architecture on AWS, Azure, and GCP, including the cost of keeping them up.
Recommendation, classification, and computer vision. Models that an account manager will adopt, measured by win rate and cycle time.
Desktop, web, and mobile products where the API contract and the screen are designed together.
A separate career assistant, grounded in this background. It answers questions about the Intuit quote systems, earlier roles, and the public agent work — with OpenAI and Gemini, and a note when someone serious writes.
People usually start with