Shalini K. Email and Phone Number

Virtualized Android Graphics Stack:Developed an advanced graphics stack to enhance performance and resource efficiency in virtualized Android environments.- Led the design and implementation of the transition to host composition, achieving a 15% FPS boost and enabling multi-window support. - Designing and implementing modifications to the Android Hardware Composer. - Designed kernel modifications to support host-side operations. - Created and optimized Vulkan-based shaders for efficient layer composition. - Implemented core Vulkan host composition logic in Rust, including a new ioctl interface for streamlined communication and data transfer. - Built a Vulkan-based gralloc service to improve performance and enable zero-copy memory sharing across processes and devices, supporting cross-process synchronization and GPU memory management. - Improved Vulkan synchronization by introducing a new kernel API for importing sync files, reducing frame overhead by 1-2ms.- Implemented robust security features to prevent unauthorized image modification.Established comprehensive CI pipelines in collaboration with the engineering productivity team to ensure stability across devices.- Partnered with AMD and Intel to develop key kernel APIs and enable external format support for Vulkan gralloc.Visual Consistency Metric:Created a new metric that adjusts dynamically to varying frame rates (beyond 60 FPS) and penalizes outlier frames, surpassing the previous Smoothness metric.Performance Analysis:Collaborated with UXR on a jank study, designing tests to identify and address bottlenecks.Built a framework for user retention analysis based on graphics metrics, simplifying the exploration of factors driving user churn.

AR/VR Camera Image Algorithms- Worked on parallel audio/video encoder pipeline to improve p75 video latency by 380msMentoring 3 new hires on the team by helping them onboard on the domain, scoping projects for their ramp up. - Conducted video compression educational classes for the broader camera teams- Working on image quality algorithm for dewarp

Maneuver Planning- Developing a test harness to test the planning stack across parameterized geometric map features to enable swift validation of planning features. - Developing metrics for the planning stack to quantify and evaluate how well the AV simulates human driving.Embedded Systems- Worked with cross disciplinary teams like ES FPGA, Perception, Hardware, Calibration, AV Frameworks and Fleet Reliability to co-design the camera data pipeline and tools for debugging and tuning the camera pipeline.- Architected camera configuration framework for the camera pre-processing module (SoC) for all FPD link cameras including deserializers/serializers/sensors/ISP configuration.- Developed drivers for Lidars (Ouster) and DMA NIC Camera driver to ingest sensor data and move data for consumption by Perception- Worked with the AV Frameworks team to propose a new design for embedded system nodes interacting with sensors to improve testability and diagnostic tools.

Transcoder ABS team- Developed a highly scalable Video Transcoder to transcode live rtmp stream into /fmp4 for HLS in real time.- Responsible for architecting features including: sprite sheet generation module for live streams, segment size analysis for ts/fmp4, new codec support for VP9.- Architected a live VP9 solution for Twitch to achieve 25% bit rate savings over H264. The goal of this project was to achieve better compression and QoS for live video, with similar latency as Twitch’s existing H264 pipeline.- Implemented end to end live multithreaded VP9 transcoding pipeline (Xilinx and NGCodec hardware).Designed frame buffer management to feed FPGAs at 60fps, added support for fragmented mp4 muxing (matroska to fmp4), CEA-608 H264 captions to WebVTT closed captioning in fmp4, software VP9 solution for transcoding ads.- Designed test plan and specifications for automated test suite for VP9 pipeline to validate and stress test new transcoder releases. Automated benchmarking using PSNR, SSIM, VMAF(using FFmpeg). Performed qualitative/quantitative analysis of videos to tune encoder settings.iOS team- Developed the iOS application for Twitch as part of a two person team, serving as a tech lead for several features including mobile broadcasting, video, chat, social features.- Designed and implemented a multi threaded native rendering pipeline for Twitch chat to improve performance, resulting in a 4% increase in minutes watched on iOS and reduced memory usage per chat initialization by 10MB.- Wrote a ​technical blog post​ discussing the multithreaded rendering pipeline - https://blog.twitch.tv/en/2016/07/15/high-performing-native-chat-on-twitch-79c1492eca06/- Introduced unit testing in the c++ SDK consumed by Android, iOS and tv apps.

WithMe Messenger App - Working on server side and front end for a messenger application.- Wrote an internal OSX app for building SpriteKit scenes from a JSON description. The JSON format includes the description of texture nodes, particle nodes, animation types and frames and other information needed to construct a SpriteKit Scene. - Wrote a push server in Haskell to integrate with Apple’s push notifications services.WithMe Ipad App - Working on the server side team for an iPad avatar based social application in Haskell and uses Redis .- Wrote a post processing component for service responses to optimistically pre-fetch related data to reduce latency.- Customized middleware for the haskell web framework (Yesod) to record API statistics. Built a Fake Yesod Handler abstraction for our custom testing framework using State Monad.- Wrote purchase services in Haskell to integrate with Apple’s in-app purchases.WithMe - Worked on a social, interactive avatar based gaming platform.- Worked on the front end of a pictionary game on this platform. Built the backend for chat.- Wrote data modelling abstractions through custom Backbone.js objects such as composite and reaction models and collections.ARM team - Full stack development in PHP using MySQL and Redis as the data storage.- Implemented the backend for a daily login reward system feature. - Added new payment channels, designed admin tools for managing promotions.Visual Stream Team - Worked on the front end of a pinterest inspired visual stream, which showcases all the user generated content and art within IMVU.

- Developed a photo management and sharing application (full stack).- Extended the people search mechanism to include location information using the Google Maps API. Used Redis as a caching layer to improve speed.