Feb 2025 – Present · 1 yr 1 mo

Dec 2023 – Mar 2025 · 1 yr 3 mos

• Conceptualized, developed, and deployed the Designer Rendering service on Kubernetes, with Node.js within a sandbox environment managed by a C# Manager. This critical infrastructure supports rendering of images and designs edited in Co-pilot and Designer, handling over 10 million requests daily with a QoS of 99.9%.
• Created a canvas wrapped in headless-gl for rendering of designs with images, videos, and graphics.
• Established a master-slave cluster architecture, with each child acquiring rendering resources from a shared pool.
• Benchmarked the service on Linux's CPU rendering virtual buffers via xfvb and mesa.
• Published gl-with-warp to leverage warp (DirectX) on Windows containers for rendering static designs, mitigating GPU costs and achieving 10-15% faster CPU rendering.
• Optimized memory efficiency of the Image Renderer creator (Presenter) for the node service, considering the low probability of re-rendering on the server.
• Implemented robust telemetry, performance logging, and distributed tracing using OpenTelemetryExporters integrated with AMACoreAgent.
• Automated memory leak detection, fidelity checks, and concurrency capability assessment in the CI pipeline.
• Developed an internal package for fidelity checking.
• Optimized passing of base64 data to reduce latency by 30%.
• Improved fidelity and implemented SVG scaling in Node.js to ensure high-resolution output when rasterized.
• Contributed to SSL termination, AKV integration, and health monitoring system.
• Sandboxing DRS:
• Utilized the MSIX and wsudo to determine capabilities of node in sandbox.
• Employed a native sandbox launcher and developed a .NET7 service to manage sandbox workers (rendering service) for isolation.
• Leveraged Yarp to proxy requests over named pipes to the worker and reverse proxy for telemetry transfer via C# service.
• Implemented workaround for firewall issues in Windows Docker containers, ensuring reliable startup of worker.

Sep 2023 – Dec 2023 · 3 mos

• Audio Export
• Added ability to export designs with audio using AAC codec from libav / ffmpeg bindings with support of filters like trim, fade-in, fade-out and volume.
• Identified an issue with aac file formats having 0 durations and implemented fixes in the bindings to handle such scenario and published them internally.
• Offline support with Local Rendering
• Implemented a http-interceptor using service worker and created asset store in IndexDB and to prevent network calls to service and render assets 30% faster with 99.9% QoS.
• Multipage Thumbnail Performance Improvement:
• Implemented support to rendering video thumbnail instead of rendering video frame on static export.
• Improved performance by 30% by unblocking main thread and making render and save asynchronous with a P50 of 50ms with 99.9% QoS with 10M thumbnails a day.
• Introduced caching of thumbnail with hash (calculated in sub-milisecond time) on page data for invalidation.
• Export Improvements
• Created an asset download manager to ensure and retry download for assets before rendering final design.
• Created a process management system to ensure rendering of asset happens only one at a time.
• Detected that some devices took away GPU resources if process was too long, added handling for 99.x% reliability.
• Improved PDF reliability on chrome version < 88, to achieve 99.99% QoS.
• Chunking in Asset download
• Implemented chunking support during asset download for faster streaming and ensuring Safari support to help 500K sessions every month.
• Misc
• Integrated GPT into a node service for demo to Satya and leadership.
• Improved caching by using URLs instead of asset ID incase of concurrent network calls.

Aug 2023 – Sep 2023 · 1 mo

• GIF Optimization for LinkedIn:
• Developed and published the groundbreaking TypeScript package, gif.js.optimized2, introducing memory-optimized GIFs up to 94% smaller than competitors. Utilized innovative algorithms including encoding transparency difference and Box offsetting, alongside global palette usage, ensuring exceptional reliability at 99.9%.
• Released gif.js.turbo, an even more optimized version boasting minimal runtime memory usage, 2-3x faster export, and 15% faster memory performance. Leveraged transferables and algorithmic enhancements to eliminate storage of uncompressed data.
• Enhanced GIF quality through GPU-based pre-dithering of frames, ensuring superior visual fidelity.
• Successfully rolled out this feature to a substantial portion of LinkedIn users, impacting approximately 9 million individuals.
• Export at 4K:
• Enabled seamless rendering and export of 4K designs without latency issues, benefiting over 100K users monthly, ensuring smooth and efficient workflow.
• Motion Thumbnails:
• Implemented the capability to preview animated suggestions on clients' machines, achieving an average FPS exceeding 20 with minimal buffering.
• Significantly reduced memory usage by 100% through texture optimization, catering to displays like Retina Display with specialized handling.
• Addressed component unmount and garbage cleanup issues, preventing memory leaks and ensuring stable performance.
• Utilized Resize Observer to dynamically scale motion thumbnails during resize operations, maintaining optimal quality and user experience.

Apr 2022 – Aug 2023 · 1 yr 4 mos

• Achievements:
• Recognized with CVP award for "Send to Phone via QR code",.
• Led Editor integration into Publish flow, winning "Best solves customer problem" in OLI org.
• MP4 Export:
• Implemented MP4 export functionality utilizing h264 codec via ffmpeg-bindings, achieving a 30% performance boost through multi-threaded encoding with libav-binding.
• Enhanced performance by 30% through webcodecs utilization for stitching frames.
• Achieved a 99% QoS on web platforms, with added features like share to phone and copy video URL, delivering seamless user experiences.
• Attained record-breaking export times on M1 chip, with a 10-second design export under 8 seconds, setting a new standard in browser-based motion exports.
• Resolved mobile latency issues, reducing latency by 3x and ensuring optimal performance across devices.
• Published npm packages of ffmpeg bindings, enabling compliant MP4 export functionalities.
• Export:
• Optimized client-side export rendering algorithm, boasting a P50 response time of 300 ms and a 99.99% QoS, establishing industry-leading static export speeds.
• Implemented Send to Design via QR and Copy as Image functionalities, for high user satisfaction with over 100K daily uses and a 99.9% QoS.
• Enhanced performance by moving rasterization to a web-worker, improving efficiency.
• Implemented robust error handling and modal UI enhancements, ensuring high reliability across browsers.
• Achieved 99%+ QoS across all browsers through meticulous bug fixes and optimizations.
• Publish:
• Developed APIs to enhance user profile integration, resulting in personalized suggestions and enriched UI experiences.
• Ensured responsiveness by implementing success/failure modals and comprehensive automation tests.
• Doubled performance with Azure blob URLs for assets, enhancing share speed.
• Branded QR Code:
• Developed a branded QR Code, featuring designer logo and colors, elevating brand visibility and recognition.
• Hack:
• Developed PoC for notification system.

Jul 2021 – Apr 2022 · 9 mos

• Achievements
• Participated and won "The Best Overall Project for Designer" in Microsoft's Hackathon, by enhancing export capabilities with Upload from phone and Send design to phone with QR code functionality (now part of multiple Microsoft products).
• Rendering Service
• Spearheaded the Proof of Concept (PoC) for a Rendering service using webpack, TypeScript, and Node.js, achieving file conversion (PNG, PDF, JPG) directly on the server.
• Optimized Rendering API for a 5x performance boost by implementing caching and pre-processing shaders, resulting in enhanced server readiness and response times.
• Implemented comprehensive monitoring solutions utilizing Grafana dashboards and Prometheus for real-time performance evaluation, ensuring service reliability.
• 3-P OAuth for Sharing Designs:
• Implemented generic OAuth C# service to help sign in using Facebook, Instagram, LinkedIn, Google or Dropbox. This was Microsoft Office's first 3-P OAuth initiative. It has enabled sign in for more than 100K users with a QoS of 99.9%.
• Implemented UI for OAuth and account selection (if multiple accounts are present).
• Added support to link / unlink multiple accounts and caching to prevent multiple sign-ins.
• Share to social media
• Added C# APIs to publish user designs to social media platforms like Facebook, Instagram and LinkedIn.
• Implemented UI for social media post creation with the addition of showing AI caption suggestions and AI hashtag recommendations.
• Watermarking
• Added support to watermark designer documents.
• Built Export UI and Polished it:
• Designed and polished a user-friendly export interface, leading to nearly 1K export requests within the first 15 days after launch of beta, maintaining a 100% reliability rate.
• Introduced PDF support with lazy loading techniques, optimizing bundle size and user experience.
• Learnings:
• TypeScript, Node.js, React, C#, and OAuth, security considerations and compliance standards within Microsoft.