Experience

Trupeer.ai

Every initiative here was built and owned solo.

Trupeer turns screen recordings into polished product videos and guides with AI. I joined to own the frontend and ended up owning everything from the design system to the backend, using agentic coding to move at the speed of a small team.

1. feat Revamped the entire product UI 45 days

   Problem

   The product had grown screen by screen, with no shared components, so every new feature looked and behaved a little differently.

   What I did

   Built a design system and component library from scratch, then moved the whole app onto it.

   Outcome

   A consistent product in my first 45 days, and a base that new features now start from instead of reinventing.

2. refactor Re-architected the API layer 40% → 95% error coverage

   Problem

   Requests, auth and error handling were spread across the codebase, so failures were inconsistent and many never reached Sentry.

   What I did

   Centralized every request, auth flow, success path and error path into one layer, then rebuilt Sentry reporting and logging on top of it.

   Outcome

   Sentry error coverage went from 40% to 95%, so problems surface before users report them.

3. feat Revamped the core video editing experience

   Problem

   The editor is where users spend most of their time, and it had to feel as fast and predictable as a desktop tool.

   What I did

   Reworked the editing experience end to end and shipped AI-generated process-flow diagrams from uploaded videos, with version history for later edits.

   Outcome

   A calmer, more capable editor, plus a new way to turn a recording into documentation.

4. ci Enforced the design system across repos

   Problem

   A design system only helps if every repo actually uses it.

   What I did

   Added robust rules, tests and CI checks that catch off-system styles and components before they merge.

   Outcome

   Consistency that holds without anyone policing pull requests by hand.

5. feat Built a Figma-to-code agent harness 98–99% accuracy

   Problem

   Turning Figma designs into components by hand was slow and drifted from the design in small ways.

   What I did

   Built an agent harness with a visual feedback loop: it generates the component, renders it, compares it with the design and corrects itself.

   Outcome

   Reusable components at 98–99% visual accuracy, and much faster UI development across the team.

   LinkedIn post about it

6. test Live testing and a post-release sanity suite 0 P0 · 0 P1

   Problem

   There is no dedicated QA team, so quality had to be built into how I ship.

   What I did

   Built a live-testing agent skill that verifies each change in the running app, and a sanity and smoke suite that sweeps every page after each release.

   Outcome

   Zero P0 and zero P1 incidents while owning frontend and backend delivery.

• React
• TypeScript
• Design systems
• Sentry
• Figma
• Claude Code
• Agent skills
• MCP

Mastercard Data & Services

Team work: I co-led the UI and QA teams.

Four years on data-intensive analytics products for banks: a founding engineer on a brand-new anomaly detection platform, and a core contributor to two platforms that won Mastercard's CEO Award.

1. feat Founding engineer on anomaly detection

   Problem

   Mastercard had no near real-time way for banks to spot unusual transaction patterns.

   What I did

   Joined as a founding engineer and built the data-intensive insights UI for analytics and alerting on banking transactions.

   Outcome

   Mastercard's first near real-time anomaly detection platform.

2. feat Core contributor to two CEO Award-winning platforms $50M opportunity

   Problem

   Two global platforms for banks, built by large cross-functional teams on tight timelines.

   What I did

   Contributed UI features to both as part of large teams, working closely with product and backend engineers.

   Outcome

   Both won the CEO Award, Mastercard's highest honour, unlocking a $50M revenue opportunity projected to drive $100M a year.

3. refactor Fake-backend and micro-frontend architecture ~30% faster delivery

   Problem

   UI work kept waiting for backend APIs to be ready.

   What I did

   Led the move to a fake-backend architecture that decoupled the UI from API readiness, and maintained a micro-frontend setup for hosting parent and child apps.

   Outcome

   Features shipped about 30% faster, and modular parent and child apps could live in one shell.

4. test Data-driven QA −70% manual testing

   Problem

   Setting up test data by hand made every QA cycle slow.

   What I did

   Automated test-data setup with PL/SQL DML scripts.

   Outcome

   Manual testing time dropped by 70%, saving 5–6 days per release cycle.

5. perf Co-led UI and QA teams of 9+ +50% performance

   Problem

   A large team needed shared standards to ship quickly without rework.

   What I did

   Co-led the UI and QA teams alongside 7+ backend engineers and PMs, ran stress and load testing, and tightened PR reviews and the release process.

   Outcome

   App performance improved by about 50%, and rework and bugs fell by about 30%.

• React
• Redux
• TypeScript
• Webpack
• Micro-frontends
• PL/SQL
• Python
• JMeter

National Institute of Technology, Warangal

Where it started: engineering at one of India's top NITs, and my first React apps on the side.