My last year at Chargebee has been about turning a single assistant into a platform other teams — and now merchants — build on, then pointing that platform at a new product surface.
AI Agents Platform · Chargebee
Built the internal platform every AI feature at Chargebee runs on, plus Backstage — the control plane for managing, evaluating and observing agent behaviour, down to per-agent latency and token consumption on every task.
The platform is now externally consumable. Chargebee’s MCP server is published in the Claude marketplace, secured with full OAuth 2.0 (pre-registration, refresh tokens, per-agent client credentials, user-level consent), and merchants can compose custom MCP servers from our toolsets. 100+ live sites run on it, adopted organically with no outbound push.
Copilot · Chargebee
Re-architected Copilot from a RAG assistant into a fully autonomous, agent-to-agent system. The main agent acts as an orchestrator, routing across specialised agents and holding the bar on resolution quality. It resolves L2+ queries automatically — the class of ticket that previously took days of human investigation.
It became a platform rather than a feature: Copilot Skills were decoupled from MCP servers, permissions moved onto the user’s own RBAC, and any team can contribute skills. Now embedded across every Chargebee app.
Billing Desk · new vertical
Leading a new receivables product — taking the AR domain expertise I built over years of ERP work and rebuilding it around automation rather than forms and queues.
- Receivables dashboard — 12-month outstanding and overdue position with an aging report, live for all customers
- Dunning sequence automation built as a generic workflow executor, with a draft-and-publish authoring experience
- Collector Queue on a multi-layer sync engine: Kafka events as the primary path, a reconciliation sweep for missed updates, on-demand admin sync, and failed-entity tracking with retries
- Precomputed customer aggregates refreshed daily, taking expensive queries off every list page
Platform engineering · Chargebee
Making the platform fast, and the stack easy to run
Cut job queue wait time 40–60%, with critical jobs down to sub-second, by restructuring runners and separating priority work from internal work. Moved every non-DB-backed task onto the durable jobs framework so nothing is lost on deploy, and added pick-up-delay vs execution-time observability to show where latency actually goes.
On the developer side: Dockerised every UI and backend service, so the whole stack boots in a couple of commands against any environment — plus a setup CLI, a dedicated predev CDN, and read replicas in EU and AU.
Chargebee Hackathon · two-time winner
First place in 2026 — and in 2020
2026: rebuilt how Chargebee’s engineering teams handle code management — an AI-powered workflow with skills that compress the development lifecycle.
2020: Chargebee as a merchant subscription aggregator driven by voice as the primary input — six years before the agent platform, and the same bet: the interface should adapt to the person, not the other way round.