Feb 2021 – Jul 2023 · 2 yrs 5 mos · San Mateo, California, United States · On-site

• License Plate Recognition (LPR) System: Developed a world-class LPR system, capable of handling diverse traffic scenarios such as highways, intersections, city traffic, and parking garages. Achieved 97% precision and 98% recall rate by integrating compact CNNs with AI-accelerated Ambarella CV22 SoCs.
• Object Detection Pipelines: Developed and maintained on-camera and backend services for people and vehicle detection. Enhancing edge inference pipeline and improved performance for long-tailed cases.
• Line Crossing and Event APIs: Integrated Ambarella APIs with our firmware for 100x faster on-camera inference. Additionally, I implemented ByteTrack with YOLOx-nano to obtain counts of people crossing user-defined lines, marking Verkada's first project using an accelerated chipset.

Jun 2020 – Aug 2020 · 2 mos · Pittsburgh, Pennsylvania, United States

• * Reinforcement Learning for Motion Prediction: Designed and implemented custom Adversarial Reinforcement Learning (RL) and Inverse RL modules for integration into the MultiXNet pipeline, enhancing motion prediction capabilities for self-driving Uber vehicles.

Jul 2017 – Jul 2019 · 2 yrs · Bengaluru Area, India

• Zero-Shot License Plate Recognition: Developed a novel image-to-text matching approach using a "symmetric triplet" loss function and a hybrid Fisher Vector + MLP architecture. This innovative solution outperformed VGG-16 with greater accuracy, 200x speed increase, and 8x less disk usage, achieving 99.6% matching accuracy on a dataset of 50,000 real plates and 96% on a simulated text database with 5 million entries.
• Vehicle Occupancy Trends System: Engineered and trained CNNs for the carpool-lane violation detection systems in the San Francisco Bay Area and NYC, enhancing detection accuracy by 5 percentage points to 95% and processing over 15 million vehicles annually. Utilized state-of-the-art models, including YOLOv3, GoogLeNet, DenseNet, and ResNet, to establish a robust system for vehicle occupancy analysis.

Jun 2016 – Apr 2017 · 10 mos

• * Established a student-led platform to document events, interview people and provide a multimedia platform for students at IIT Delhi, while securing university approval and funding for the project. Successfully built a dedicated recording studio and founded the first 10 member strong OnAir team. Our YouTube channel has over 33k subscribers today and continues to provide a platform for the student community at IIT Delhi.

May 2016 – Aug 2016 · 3 mos · New Delhi

• * Analysis of Sparse Subgraphs in Large Networks: Investigated the relationship between the degree distribution of large-scale graphs and a suggested modification to the local clustering coefficient, providing valuable insights into the behavior and structure of sparse subgraphs within complex networks.

May 2015 – Aug 2015 · 3 mos · Greater Chicago Area

• Under the guidance of Prof. S.D. Joshi (IIT Delhi) and Dr. M. Kotecha (University of Illinois at Chicago), I participated in a project focused on measuring biochemical concentrations in MRI images. Utilized ML techniques such as K-means clustering and PCA for region of interest segmentation and non-invasive tissue development monitoring. Additionally, I attended the In Vivo Electron Paramagnetic Resonance (EPR) Imaging workshop at the University of Chicago to further expand my knowledge in the field.