Sep 2024 – Sep 2024 · 0 mo · Munich, Bavaria

Apr 2024 – Aug 2024 · 4 mos · Munich, Bavaria, Germany · On-site

Jul 2022 – Mar 2024 · 1 yr 8 mos · Munich, Bavaria, Germany · On-site

• Rethinking L2 systems in autonomous driving with unsupervised learning and no-label data streams

Jul 2021 – Jun 2022 · 11 mos · Bengaluru, Karnataka, India · On-site

• Devised AI solutions for complex ADAS challenges in a think tank setup,
• Scrutinized vendor PoC models by evaluating them with a toolbox that consolidated 5 handcrafted adversarial attack methods (for monocular 3D object detection)
• Attenuated the dimension & orientation error for truncated and occluded 3D objects using various dead-pixel augmentations, boosting 3.5% mAP on inhouse dataset
• Introduced road surface modelling & active lane tracking into lane detection pipeline that transformed it from 2D to 3D output delivery
• Modelled a lane network mapping algorithm using lanelets & over-the-time aggregation of highway lanes as a ground truth evaluation tool

Jul 2019 – Jul 2021 · 2 yrs · Bengaluru, Karnataka, India · On-site

• Built pre-dev versions of Android in-vehicle-infotainment system and remote control app,
• Developed an Android background service to stream internal audio (via UDP) with a max source-to-sink latency of 16ms in addition to an H264 screencast service (via RTP) with a max glass-to-glass latency of 70 ms
• Removed 20% of manual UAT backlog per cycle by articulating a Rig-in-Loop testing procedure for UAT automation of JLR mobile app
• With smart pointer dereferencing and clean array allocation in embedded C, reduced the peak CPU utilization of suspension control firmware by 2.8%
• Co-developed a talent assessment web-tool (based on LAMP stack) with HR department, saving $9000 in subscription cost on third party

May 2018 – Jul 2018 · 2 mos · Bengaluru Area, India · On-site

• Contributed towards the development of a traffic surveillance system product,
• Ensembled a state machine of 3 Neural Nets, operational on thermal camera, that furnished vehicle count statistics for toll booths with a 95.9% recall
• Incorporated a depth injection method based on single-view geometry to improve the wheel size regression, thereby boosting mean recall scores by 2.5%
• Refactored the legacy codebase using OpenCL & CUDAbindings of OpenCV, increasing the GPU utilization and reducing the peak CPU usage by 12.5%

May 2017 – Jul 2017 · 2 mos · London, United Kingdom · On-site

• Blue vision labs was acquired by Lyft Level 5 which then later joined forces with Toyota Research Institute
• As part of perception team,
• Setup an offline structure-from-motion baseline that reconstructed dense road surface at city-scale (San Francisco) using crowdsourced mobile camera images
• Built a flask web tool, hostable on Amazon MTurk & capable of tagging road surface artifacts in BEV space, that helped annotate 9K samples within 2 week

Apr 2016 – Aug 2016 · 4 mos · Department of Mechanical Engineering · On-site

• Supervised by Prof. Dilip Kumar Pratihar
• Developed an exoskeleton structure for lower extremity of physically disabled to help impart locomotive ability
• Implemented a wireless network of RF modules consisting of IMUs to record GAIT cycle data in real time
• Fabricated a plantar system to measure ground reaction forces at crucial points in foot sole in a GAIT cycle

Mar 2015 – May 2019 · 4 yrs 2 mos · IIT KHARAGPUR · On-site

• Autonomous Ground Vehicle (AGV) is research group established with a direction to produce high-quality implementations for Autonomous systems like self-driving cars.
• The team also participates every year in Intelligent ground vehicle competition (IGVC) wherein the challenge is to produce an autonomous robot traversing unknown paths and terrains.
• As a team head for my time in AGV,
• Led the team towards second podium finish at IGVC'18
• Help team reach within top 14 teams for Mahindra Rise Prize challenge
• Acted as enabler for fellow students to publish at relevant conferences
• As an engineer,
• Developed a lane detection algorithm using classical techniques like SVM and inverse perspective transform
• Implemented a traffic sign recognition system using histogram of oriented gradients matching
• Implemented a monocular visual odometry pipeline using an end-to-end neural network approach
• Designed & fabricated the electronics framework for in-house robots like BMS, actuator controls & telemetry
• Programmed ROS packages for steer-velocity control and perception stack of the vehicle