Jan 2022 – Sep 2023 · 1 yr 8 mos

• Predicted intents for Waymo Driver using machine learning and reinforcement learning techniques.
• Developed and launched Waymo’s first closed loop learning system in planning. (Work recognized by spot bonus and pending patent)
• Led ideation and mentorship for several planner RL formulations which nurtured further adoption and generalization of the built system.

Jun 2018 – Jan 2022 · 3 yrs 7 mos · Mountain View

• Joined as part of founding ML team.
• Built ML components based on 3D Vision and Reinforcement learning for multi modal perception and planning system

Jan 2018 – Mar 2018 · 2 mos · Greater San Diego Area

• Served as a Teaching Assistant for course on Neural Networks under Prof. Garrison Cottrell.
• Course Link: https://podcast.ucsd.edu/watch/wi18/cse253_d00

Sep 2017 – Dec 2017 · 3 mos · Greater San Diego Area

• Served as a Teaching Assistant for course on probabilistic inference on graphical models under Prof. Lawrence Saul.
• Course Link: http://cseweb.ucsd.edu/classes/fa17/cse250A-b/

Sep 2017 – Jul 2018 · 10 mos

• Contributed to projects related to social trajectory predictions for surround agents and learning based solutions using geometry and deep learning for 3D object detection and tracking for autonomous vehicles under the guidance of Prof. Mohan Trivedi.
• Eventually lead to publications at ECCV (2018) and ITSC (2018).

Jun 2017 – Sep 2017 · 3 mos · Palo Alto

• Working on computer vision and deep learning technologies at A9's Visual Search group.
• Key Points:-
• 1) Fast 2-D Semantic Detectors for Mobile Subsystems
• 2) Semantic Segmentation
• 3) Design Evaluation Framework

Jul 2014 – Jul 2016 · 2 yrs

• I was a part of Multimedia Research team in the Broadband and Connectivity Group and worked on the design and implementation of optimized computer vision algorithms for Advanced Driver and Assistance Systems. I was involved in the complete analysis of the algorithms involving hardware constraints for low power SIMD cores to make them perform at a processing speed of 30fps . I also worked on the Video Firmware Design and implemented various video encoding modules for Set-Top-Box Chips.
• Following were the main projects on which I worked at Broadcom:-
• 1) Developed proprietary computer vision applications like Pedestrian Detection, Traffic Sign Detection, Lane Departure Assistance, Face Detection as a part of ADAS system
• 2)Devised new implementation techniques to avoid iterative computations on scaled down images using trained SVM classifiers for applications like Pedestrian Detection
• 3)Implemented code organization techniques and data memory access optimizations which quadrupled the frame processing capabilities for vision applications
• 4)Implemented Core Functionality of VP8 Encoding Format for enabling the video conferencing feature on Broadcom's STB Platform. Designed a pipeline mechanism to integrate different Hardware Blocks to accelerate the process of encoding. Developed a Frame Level Rate Control Scheme to manage the bit rate as a function of resolution.
• 5)I worked on the development of Multimedia Algorithms on Low Power Smart Watches in the IOT Segment. A fully optimized MP3 decoder was written in ARM Assembly to accelerate the functionality on ARM Cortex M4 processors.

Jan 2014 – Jun 2014 · 5 mos

• I was involved in the development of Dolby Multistream Platform on the Broadcom's audio subsystem.It is a single package solution for all audio contents which are used worldwide on the Set-Top-Box Platform. My main contribution was towards the Dolby Digital Plus Encoding which is a transcoding feature used to produce 7.1 audio bitstreams. A fully optimized algorithm was developed which was coherent with all Single Input Multiple Data Instructions used in the Audio Processor. To speed up the DMA optimization procedure, DMA profiling was automated using the knowledge of Perl.The butterfly structure of FFT was coded in assembly to generate better results than assembly generated code.

May 2013 – Jul 2013 · 2 mos

• Developed a Parachute Simulator System with proper hardware interfacing using LABVIEW. The software test suite developed was tested with a simple hardware implementation of an I to V converter on PCB and photo-detector, pressure and wind sensor interfaced with the system. The project got recognition in the defense research for it's help in providing information in emergency situations about the auxiliary parachute attached ,wind passage and starting threshold speed.

May 2012 – Jul 2012 · 2 mos

• Worked on a Project related to Relay Communication System for Power Distribution in AC Auxiliary Plants , The project involved an overview of all the basic components involved in the communication system using Data Concentrators.