Aug 2016 – Feb 2022 · 5 yrs 6 mos · Bengaluru Area, India

• Part of the XR Research, leading graphics research projects on new ideas for Qualcomm's Boundless XR initiative.

Nov 2013 – Jul 2016 · 2 yrs 8 mos

• I am leading the Virtual and Augmented Reality effort within Samsung Bangalore. I am also involved in building tools and pipelines for migration of existing OpenGL software to the Vulkan API and Samsung's Vulkan CTS effort.
• I was involved in the creation and feature development of the open source VR library - the Galaxy VR Framework Library (www.gearvrf.org). I worked with my team to add core Scene Graph features such as hierarchical model loading using Collada, Frustum Culling, Occlusion Culling, Remote Rendering and other features. This activity was done in collaboration with Samsung Research America (San Jose)
• Earlier, I worked on building the Effects Engine Framework - an engine and associated pipeline to allow easy loading of large 3d models and animated characters on mobile devices using the Khronos Collada Format. This work was published in Siggraph 2014 as a poster. I also worked on building a prototype AR/VR Switching system and framework used for a multiplayer AR Chess game. This was published in Siggraph 2015 as a poster.

Apr 2009 – Nov 2013 · 4 yrs 7 mos

• Worked as a Technical Director for Animation and Lighting. Credited in the Oscar Nominated feature film - Puss in Boots, the huge hit Madagascar 3 and its sequel The Penguins of Madagascar. My primary job involved developing and maintaining the extremely complex DreamWorks Animation pipeline and working day to day with artists in solving their problems and fixing bugs. I was completely responsible for the deployment and implementation of the current generation Animation Pipeline at the Bangalore studio. I spent over three months at DreamWorks Animation Redwood City working with the TDs there to build, understand and ease this transition to the newer animation pipeline in India. The new generation pipeline was used for Animation in The Penguins of Madagascar.
• I also did the following major project in spare time:
• Interactive Fur Preview for Emo: Emo is the DreamWorks’ interactive animation package and it did not support real-time fur rendering. Sometimes animators required to visualize the fur silhouettes to know if the fur is crashing with the body. The normal way to do this is to launch a render on the render farm and check the resultant output - a process that normally takes at least 10-15 minutes.
• I developed a plugin for emo that allowed animators to visualize fur in real time using OpenGL. The fur was geometry and color accurate. We sampled a pre rendered shaded point cloud of the geometry for color information of the fur and used preview textures for texture mapping the surface. This system allowed for instant turnaround times of artists instead of having to wait for rendered frames. Even though the original system didn't support GPU based rendering, our system supported GPU based vertex and index buffered (VBO/IBO) geometry and a modern GLSL shader system for full flexibility.

Sep 2005 – Apr 2009 · 3 yrs 7 mos

• Basic research on Large model rendering, Geometry Streaming & Virtual Navigation. Represented MSR India at MSR Techfest in Redmond in 2006, 2007, 2008. Spent large amounts of time at MSR Redmond collaborating with researchers.
• Geometry Streaming Systems: Designing a working geometry streaming system requires a combination of subsystems such as visibility culling, path prediction, caching and others. This work tries to quantify the effects of each in the performance of the streaming system based on the viewer experience and network utilization. We also explore the effects introducing dynamic objects into the scene.
• Real-time Occlusion Culling for Dynamic Environments: Occlusion Culling for large dynamic virtual worlds is a hard problem. Traditional techniques for occlusion culling utilize either a pre-computed data structure such as a BSP tree and then select portions of the BSP at runtime. While this method works exceedingly well for static entities, it fails for dynamic worlds where the need for recalculating the BSP upon object motion can’t be achieved in real-time. We propose a system which utilizes a combination of a loose reconfigurable hierarchy, occlusion queries and rearrangement of the hierarchy at runtime to improve performance levels of occlusion culling systems in dynamic scenes.
• Assisted Navigation in Virtual Environments: Navigation in large virtual environments is a well studied problem. A significant majority of techniques used for navigation use some form of developer generated waypoint systems with traditional beam or A* algorithms. In this project, we use a-priori knowledge of previous user track-logs to improve the performance of existing navigation algorithms. We apply the algorithm in two distinct domains – Virtual Earth 3D(GIS) to find interesting areas on a 3D map and in a computer game for AI Bot navigation. The significant advantage of this method is the fact that it can be combined with traditional approaches for better performance.