Agent Orchestrator
A multi-agent orchestration system from 2026: three specialised agents and an orchestrator, built with LangChain and the Claude Agent SDK. It is the internal reference implementation for agentic AI at the firm.
Work
Ten years from test automation to agentic AI, with five years of modernization at scale in between. The thread: build the harness that lets a team trust automation.
Watch the 40-second film
Career at a glance
career.yml
5 of 6 stages passed, 1 running
DollarsGPT passed
- Founded a consumer rewards platform while at university
- Grew from zero to 20,000 users with 1,500+ monthly actives
- Ran development, partners, payments and growth
Infosys passed
- BDD frameworks with Selenium, Java and Cucumber
- Clients: Telstra and the Government of India
- A report module that made the team 11 times more productive
Test automation passed
- An organization-wide test automation framework at Morgan Stanley
- Email-embeddable Cucumber reporting for QA teams
- Co-invented BLAZE: API specs in, Java test suites out (US patent)
Tech at scale passed
- Five years modernizing technology across Morgan Stanley business units
- Engineer as a Service: 35% faster modernization rollouts
- Led 5+ teams to a 40% jump in deployment frequency
- Cookie-cutter patterns and Kubernetes adoption: 28% faster time-to-market, 45% fewer incidents
Agentic AI passed
- VP, AI Platform & Tech Modernization
- Led agentic AI adoption and literacy across engineering teams
- India Technology Showcase Award for the ACT Conference Portal
In progress running
- Agent Orchestrator: three specialised agents under one orchestrator, on the Claude Agent SDK
- Building agents that ship safely inside a regulated enterprise
- Leaving a cookie-cutter tool behind every fix
What I built at Morgan Stanley
These are internal systems, so there are no screenshots or code here, only what they do.
BLAZE (US patent, co-inventor)
BDD Lean Automation Zenith Engine: turns API specifications into executable Java test suites using behaviour-driven development. Granted as US Patent 12,321,256 B1, "Behavior data driven automation test code generating system and method", on 3 June 2025, assigned to Morgan Stanley.
Internal events and conference platform
Built and evolved over 5+ years into the standard for global internal events. Added AI features: proposal submission and review with a recommendation engine for session curation, content filtering, and a Hacker News-style knowledge platform with AI ranking.
AI developer tooling
A VS Code extension for in-place codebase review and modernisation, CLI skills for DevOps automation, and reusable skills infrastructure that cut onboarding and deployment tasks from days to minutes.
Pull request chat
A chat bubble on the pull request page so authors can talk to reviewers in place, saving 375+ person-days of effort a year.
Every role with dates and all its highlights is on the CV.