Sep 2023 – Nov 2024 · 1 yr 2 mos

• Architected and implemented a high-throughput service that streamlines data retrieval for LLM models like BERT, T5 and GPT from SQLite.
• Optimized the inference pipeline by refactoring critical components in C++ (CUDA, ONNX, and TensorRT) for accelerated processing, reducing overall latency by 40%.
• Conducted multi-GPU, Distributed Data Parallel (DDP) training jobs and designed dashboards in AWS Sagemaker, for efficient hyperparameter optimization and model interpretability.
• Integrated OpenTelemetry and Datadog across the stack for improved infrastructure monitoring and visibility

Jun 2022 – Sep 2022 · 3 mos · Santa Clara, California, United States

• Developed and piloted an end to end real-time Video Object Segmentation (VOS) framework for user prompted target segmentation and tracking in video calls. Finetuned a lightweight U-Net model to achieve instance segmentation accuracy of ∼ 92%
• Performed inference optimizations to establish guardrails for memory usage < 10MB, while simultaneously achieving inference speeds of ∼ 40 FPS
• Ported the model to C++ (ONNX) and performed cross platform analyses on various CPU and GPU hardware

Jan 2022 – Jan 2023 · 1 yr

• Teaching assistant for the course CS220 - Data Science Programming

Jan 2022 – Jun 2022 · 5 mos

• Co-led the Sensing and Prediction Research Group as part of UW-Madison’s Collaborative Autonomous Driving (CADS) research team
• Performed in-depth ablation studies on modern architectures such as YOLOv5 and Mask RCNN for optimizing the trade off between speed and accuracy
• Integrated the simulation software’s API with the testing pipeline to facilitate Continuous Integration (CI) and seamless inference

Sep 2021 – Dec 2021 · 3 mos

• Designed an end to end system that applies Deep Learning to detect and estimate falls in elderly people, in real time.

Sep 2020 – Sep 2021 · 1 yr

• Designed and maintained Extract Transform Load (ETL) pipelines for migration and integration of data into the Oracle Cloud Data Warehouse from sources like AWS S3 and Azure Data Lake. Scaled to sizes > 1 Peta-Byte.
• Designed several event trigger based workflows for real time updates and notifications
• Generated reports and dashboards by integrating with apps like BI Publisher and Power BI

Jun 2019 – Jul 2019 · 1 mo · Bengaluru Area, India

• Designed and simulated an end-to-end Autonomous Optical Navigation (OpNav) system for spacecraft navigation
• Simulated satellite orbital sensor data and images from software such as Celestia and Stellarium
• Built an image processing pipeline to clean and denoise the image data using MATLAB’s Image processing Toolbox
• Iteratively predicted satellite coordinates, in real time, using an Unscented Kalman Filter

Jun 2018 – Jul 2018 · 1 mo · Bengaluru Area, India

• Developed a novel machine learning framework for the classification of targets detected by the Through the Barrier Imaging Radar, a
• state-of-the-art radar at LRDE
• Engineered relevant feature vectors from raw intensity maps produced by the radar
• Built a Support Vector Machine (SVM) model capable of processing and classifying the extracted features