I take products from zero to scale in video conferencing

ProblemMost of Reliance's workforce of around 600,000 collaborated over WhatsApp and phone calls. JioMeet only did video meetings.

ActionI lead the 11-person team turning JioMeet into chat, audio and video in one product inside JioWorkspace, Jio's productivity suite. I own the client architecture across web, desktop, iOS and Android, co-own platform decisions with the media and backend architects, and partner with product managers on the roadmap.

ResultAbout 100,000 people across Reliance now use it actively. AI collaboration features are next.

ProblemFeatures took two to three weeks to reach users, and every release waited on a manual QA cycle.

ActionI moved the team to acceptance-test-driven development and a full CI/CD pipeline. Every feature ships to production behind a flag, gets verified there by QA and product, then rolls out to customers in phases. I also set up the sprint rhythm with product: a prioritised backlog, weekly grooming, t-shirt-sized estimates, spike tickets for unknowns, and a demo and retro every sprint.

ResultTime to ship dropped from 2–3 weeks to 2 days, and the team releases with more confidence than before.

ProblemWeb, Electron desktop, iOS, Android and backend each lived in their own repository. Nobody, human or AI, could see a feature end to end.

ActionI merged them into one monorepo with shared packages and a design system, extending the pattern I first set up at Sequence.

ResultWith full context available to AI tools, every engineer now delivers features end to end across all clients and the backend.

ProblemThe roadmap favoured advanced chat features such as message forwarding and an AI assistant. Usage data showed most users relied on calls, mostly on Android phones and the desktop app, often on weak networks.

ActionI took the data to product and argued for meetings first: recording, an automatic switch to audio-only when the network drops, a smooth mobile join, and stability. On a weak network the client now protects audio first, then screen share, and gives up video first.

ResultMany recurring complaints about meeting quality on low bandwidth stopped. The chat features followed once the core was solid.

ProblemBreakout rooms were used in only 1–2% of meetings, so product kept them at the bottom of the list.

ActionQA analysis showed the feature had basic issues and didn't work reliably. I pushed to fix those before deciding anything about its future.

ResultUsage rose to about 5% of all meetings, more than doubling, for a feature used mainly by education customers.

ProblemMeeting features were needed both inside JioMeet and by external customers building their own apps.

ActionI architected a modular client and published it as the JioMeet SDKs on npm: core, hooks, store, media, templates and plugins.

ResultJioMeet itself is the largest consumer of its own SDKs, so external customers run on the same code our users do.

ProblemA Chartered Accountant's associates spent a full day per client every month matching purchase records against the GST portal in Excel, carrying unmatched invoices forward and updating the portal one invoice at a time.

ActionWith the CA as product owner, I designed and built a multi-tenant platform for GSTR-2B reconciliation and GSTR-3B filing. Wrong tax logic fails silently, so every rule became a test before AI agents implemented it.

ResultOn real client data, a client with 500+ invoices went from a day to about 30 minutes, with every figure computed by the system. Small clients save little; the value grows with volume.

ProblemSequence automates billing, invoicing and revenue recognition for software companies. It needed a customer app, admin and super-admin portals and a mobile client, and had no frontend engineer.

ActionI joined as the first one and built the frontend from scratch as a monorepo, with shared packages for common logic and one UI component library and design system for every app.

ResultFour apps on one consistent system, and a pattern I later took to Jio. The role ended in a company-wide layoff.

ProblemWhen someone reported a bad call, engineers could only look at server logs. What happened on the user's device was invisible.

ActionI captured WebRTC stats from clients and turned them into Datadog dashboards for per-call debugging, plus trend views and a call-quality index.

ResultEngineering and support could trace a complaint to its cause and watch quality across every call. The role ended in a company-wide layoff.

ProblemJioMeet ran on a third-party media engine, which limited control over cost and quality.

ActionI built the JioMeet client for Jio's own mediasoup-based engine and worked on scaling the product. I'm a co-inventor on an Indian patent for dynamically generating SVG graphics in video conferencing, filed during this role.

ResultJioMeet moved onto its own engine. I left to move back to Pune, and Jio hired me back in 2023.

ProblemDolby built conference-room hardware and audio codecs; BlueJeans needed the meeting experience to run on it.

ActionI built and led the product with three developers and a QA automation engineer: a meeting app on Dolby's hub in a custom Chromium build tuned for Dolby audio, and a remote-control app over wired or wireless links for calendars, joining and in-meeting controls.

ResultAbout 5,000 rooms worldwide, many in executive offices, where CXOs used it as their main meeting device.

ProblemBlueJeans started as a Python/Django app. As it grew into a browser meeting client used by millions, the frontend had to keep up.

ActionAs the first frontend engineer, I led the client through jQuery, Backbone and Marionette.js to React, adopted early, and later from class components to hooks.

ResultA modern stack under a product that never stopped serving users, through to Verizon's acquisition of BlueJeans in 2020.