I build products for the agentic era.

The agentic layer on top of analytics: agents that build dashboards from a sentence, and the tooling that lets other agents do the same.

• Widget Copilot: a Java agent workflow (intent identification, module-selector guardrails, widget generation, summary) streamed token by token over SSE to a Vue chat UI with partial-JSON rendering and persisted history.
• Dashboard generation agents that produce whole dashboards and widgets (cards, gauges, charts, list views, narratives, images) from natural-language prompts, with prompt editing, thumbnails and AI licensing controls.
• Python MCP server and connectors exposing widget creation, data fetch and signed-URL exports as tools, with per-user service tokens propagated through tool calls so agents act with the caller's permissions.
• Agno orchestration in the Python AI service (widget router, intent identifier, prompt builders, copilot tools) called by the Java backend for multi-step workflows; a WebMCP tool registry with an iframe bridge so in-page agents can drive the client.
• Authored and maintain the agents' system-instruction library and the reference corpus they ground on: internal query language, layout grammar, widget-edit guides and a facilities-management language map.

Built the dashboard product from scratch and carved it out of the monolith.

• Standalone Vue 2 micro frontend: grid layouts, tabs and grouping, KPI and card widgets, module and pivot reports, live module reports, drilldown filters, timelines, sharing and permission scoping, app-scoped dashboards and translations.
• Split dashboards and reporting out of the monolithic client into an independently built and deployed module with an embed bridge, shrinking the host bundle and decoupling release cadence.
• Designed and implemented the analytics and dashboard API layers in the core Java server: report data, folder CRUD, widget lifecycle, report, pivot and dashboard export with signed URLs, permission definitions and route configuration.
• Cut the production JavaScript bundle from ~1 MB to ~150 KB through Webpack tuning, lazy-loaded grid widgets and dependency pruning.

A cross-app launcher, the identity API behind it, and an SDK host apps adopted within a week.

• Home API in the Java identity service: RBAC-backed, user-scoped app catalog, CORS filter, server-to-server client to the core app and platform-app launch URL resolution, rolled out across all deployment regions.
• React 19 + Vite launcher with per-app roles, profile, org switching, authenticator setup and confirmed destructive account actions, working around cross-site cookie and regional CORS constraints.
• Five-package SDK: headless TypeScript core, a Stencil web component, auto-generated React and hand-written Vue 2.7 bindings, and a React Native switcher, published to npm via CodeBuild CI.

Planned-maintenance scheduling, delivered end to end.

• Java backend for folder and PPM CRUD, criteria, pagination, dynamic KPIs and rescheduling; Vue + Bryntum SchedulerPro frontend.
• PDF export through the Puppeteer PDF service with timeout handling and isolated browser contexts; fixed a series of multi-timezone rendering defects reported from production.

The packages every Facilio app consumes, and the generation target for the AI agents.

• Charting library designed from scratch on Apache ECharts 5: bar, spline, pie, gauge, treemap, heatmap, telemetry-map and baseline charts, drilldowns, palettes, dark theme, unit and decimal formatting, timezone and DST handling; shipped as a versioned package.
• Co-maintainer of the design system: reading and dashboard cards, page header, cascader, accordion, steps, modals, gridstack sub-grids and the default theme tokens.
• Maintained the filter and criteria builder across two major versions with a beta-then-stable release flow; contributed to the shared forms package.
• Own the SVG icon pipeline and categorised icon set, the internal docs site with icon search and deprecation, animated icons, and publishing to the private registry.