1. Developing tech to improve the company’s Sense-and-Avoid (SAA) platform: A vision based product for Unmanned Aircraft Systems to navigate in unregulated airspace (using image streams from ordinary cameras & object detection/classification/tracking technologies). I have been involved with the following components: (i). Deep Learning Framework: - Spearheading the development of an end-to-end deep learning pipeline having the capability of detecting, classifying, and… Show more 1. Developing tech to improve the company’s Sense-and-Avoid (SAA) platform: A vision based product for Unmanned Aircraft Systems to navigate in unregulated airspace (using image streams from ordinary cameras & object detection/classification/tracking technologies). I have been involved with the following components: (i). Deep Learning Framework: - Spearheading the development of an end-to-end deep learning pipeline having the capability of detecting, classifying, and tracking aerial objects in diverse/noisy environments. - Managing datasets comprised of synthesized and real-world images as part of this effort. - Utilizing cutting-edge software and MLOps tools: AirSim (object simulator), Gazebo (3D robotics software), TFX, PyTorch, ROS (Robot Operating System), Anyscale Ray, RoboFlow, MLFlow. (ii). Uncertainty Quantification for increased Trustworthiness: - Developed a novel uncertainty quantification method within the deep learning pipeline which can be implemented in real-time. - This enhancement enabled the estimation of epistemic uncertainty associated with the model’s predictions, enabling the user to ascertain how much they should trust the model results (in real-time). Achievements: - Drastic improvement in range of object detection (by at-least 150%). - Drastic reduction in false detections in noisy environments (by at-least 75%). - Improved below-horizon detection capability (object detection with landscape as the background instead of sky) in terms of detection range. - Made a systematic pipeline that is easy to read, maintain and put into production.2. Research Proposals: Have submitted (to many agencies including US Air Force, Navy and DHS) various technical proposals (currently under review) containing novel ideas for tackling modern technological challenges using AI and computer vision to potentially acquire a total of $2-2.5 million for the company. Show less

• Developed a novel physics-inspired Uncertainty Quantification (UQ) framework capable of single-shot estimation of a neural network's prediction uncertainty by leveraging kernel methods (work published in prestigious venues: Neural Computation, UAI, etc.).• Framework surpassed the performance of state-of-the-art (SOTA) uncertainty estimation methods by 10-15% in terms of ability to detect false model predictions (measured by receiver operating characteristics, precision-recall &… Show more • Developed a novel physics-inspired Uncertainty Quantification (UQ) framework capable of single-shot estimation of a neural network's prediction uncertainty by leveraging kernel methods (work published in prestigious venues: Neural Computation, UAI, etc.).• Framework surpassed the performance of state-of-the-art (SOTA) uncertainty estimation methods by 10-15% in terms of ability to detect false model predictions (measured by receiver operating characteristics, precision-recall & correlation metrics) and by 10-15% in terms of calibration metrics (Brier score, expected calibration error) while decreasing compute time to 60% that of SOTA (Models: ResNet, VGG, LeNet. Data: CIFAR-10, MNIST, ImageNet).• Also achieved an average improvement of 10-15% over the state-of-the-art in the application of scene segmentation for autonomous vision (Models used: SegNet, FCN-8, PSP-NET, and U-NET. Datasets: CamVid and Cityscapes. Metrics: patch-accuracy, patch-uncertainty and PA vs PU scores). • Other applications where the framework performed significantly well: Classification under data distributional shifts, anomaly detection & transfer learning.• Developed a Hierarchical Linear Dynamical System (HLDS), that improved over the traditional Kalman filters in applications spanning video game action sequence segmentation (DARPA project), dynamic texture synthesis, and speech phoneme recognition. Show less

• Course: Machine Learning for Time Series (Instructor: Jose C. Principe) - Theory of adaptation with stationary signals, performance measures, LMS, RLS algorithms, implementation issues and applications.• My role included clarifying concepts and questions posed by students, grading assignments, help develop curriculum and assist in delivering lectures.

• Aventusoft LLC is a research startup that develops medical devices for high-value cardiac assessments by analyzing heart valve movements. I worked with the HEMOTAG device, the flagship product of Aventusoft for diagnosing and managing heart failure assessments. My contributions included the following:• Implemented deep learning algorithms for detecting anomalies and fiducial points/events in Electrocardiography (ECG) time-series data as part of a downstream task of arrhythmia… Show more • Aventusoft LLC is a research startup that develops medical devices for high-value cardiac assessments by analyzing heart valve movements. I worked with the HEMOTAG device, the flagship product of Aventusoft for diagnosing and managing heart failure assessments. My contributions included the following:• Implemented deep learning algorithms for detecting anomalies and fiducial points/events in Electrocardiography (ECG) time-series data as part of a downstream task of arrhythmia detection. The work was incorporated into the HEMOTAG product.• Tested and validated algorithm’s performance on benchmark ECG datasets (such as MIT-DB, European ST-T and PhysioNet). Achieved improvement of 30% (F1-score, ROC-AUC, accuracy metrics) over company's existing algorithms at the time.• Collaborated with the research team to discuss and suggest future research work to improve the HEMOTAG technology, specifically to tackle issues involving interpretability of AI algorithms when implemented on medical time-series data. Show less

• Worked at the Commercial Vehicle Business Unit (CVBU) of Tata Motors (Pune plant). • Job involved studying vehicle assembly line automation systems, suggesting improvements and carrying out major maintenance operations when necessary.• Made key technical improvements in several production automation systems with respect to safety, maintenance and productivity.• Performance rated in top 10% of all employees during the first year.