I build and orchestrate multi-agent AI systems. Ten-plus years shipping full-stack platforms across SaaS, fintech and enterprise — now designing the agent runtimes, MCP tool servers and RAG pipelines that put them into production.
Years shipping production software
Small businesses served by the platform
Duplicated UI code eliminated
Test coverage established at PayPal
I've spent ten years building the web platforms companies actually run on — CRM and marketing automation for small businesses, member eligibility at USAA, payment flows at PayPal. Front-end architecture is where I'm deepest: React, Vue 3, TypeScript, and the build systems and component libraries that decide whether a team of engineers ships weekly or monthly.
For the last two years that work has pointed squarely at AI. I architected and shipped a Python agent orchestration service — a multi-tenant API where internal product teams register, configure and run their own LLM agents — replacing the per-team one-off integrations that came before it. Around it I built MCP tool servers, a memory RAG pipeline on a vector database, and the in-product AI chat experience end to end.
I lead onshore and offshore engineers from architecture through Kubernetes deployment, and I care a great deal about the unglamorous half: distributed tracing, instrumentation, build times, and removing the duplicated code that quietly slows everyone down.
The agent orchestration platform I architected at Thryv — a multi-tenant service where product teams register and run their own LLM agents. Hover any node to see what it does.
Washington State University
May 2020Bhilai Institute of Technology, Raipur
June 2016AWS and React — Creating a Full-Stack Application (LinkedIn Learning)
LinkedIn LearningA multi-tenant Python API where internal product teams register, configure and run their own LLM agents — replacing the per-team one-off integrations that came before it across a CRM and marketing-automation platform serving 15,000+ small businesses.
Model Context Protocol servers giving agents governed access to product data and internal APIs. Cut the effort to add a new AI integration from weeks to days.
A retrieval pipeline on a vector database that grounds agent responses in company data — embeddings, retrieval and conversation memory, so answers cite the business rather than the model's priors.
The end-to-end chat experience: streaming responses, conversation state, retry and fallback handling — built in Vue 3 and React against the orchestration service.
Platform-wide front-end architecture — a shared Vue and React component library plus the Webpack-to-Vite migration. Cut CI build times ~30% and removed ~40% of duplicated UI code.
React and TypeScript platform determining member qualification for USAA banking and insurance products, serving a member base in the millions. Route-level code splitting cut initial bundle size 15%.
Single-page payment flows in React and Redux with JWT authorisation and Node.js data-exchange layers — plus the build pipeline and test suites that took coverage to 97%.
A MERN-stack visualisation application for anomaly detection on electrical grid data, published as the CP-SAM security assessment tool at Washington State University.