Edge AI Accelerators and Scalable Machine Learning at the Distributed Research on Emerging Applications & Machines Lab (DREAM:Lab), Department of Computational and Data Sciences (CDS), IISc.

Contributed to the development and launch of a platform that integrates real-time IoT and drone data with AI, AR/VR, and generative AI, aimed at enhancing urban infrastructure management and smart city planning. Updates :• Selected for the IIT Guwahati Technology Incubation Centre, Innovation Catalyst Program and completed it from May - July 2024. Selected for Incubation at IITG - TIC and Entrepreneurship Development Program (EDP) by Government of Assam and IITG Research Park.• Selected for the Falling Walls Lab - Tokyo 24’, IIT Bombay TIH-IoT: EIR Program [R1] in July 2024.

Served as an AI Advisor for DigiLab (DS-2) at the State Office for Roads, Bridges, and Waterways (LSBG), Government of Hamburg, Germany. Contributed to developing deep learning models in PyTorch for pattern recognition and exploratory data analysis on various sensor data (time series, geolocation, LiDAR). Deployed prototypes with a microservice architecture using Flask API, Docker, and k9s for MLOps.• Received recognition from Hamburg's Senator Dr. Anjes Tjarks and LSBG CEO Dr. Klotz for developing a drone control station server and implementing AI features like hand gesture recognition and SLAM. [ live demo presentation (April 2023) : https://bit.ly/3XHB3fw ] Tech Stack: Python, PyTorch, TensorFlow 2.0, Flask, MapboxJS, Docker, ROS 2, Kubernetes, C++, Kafka.

• Conducted research on ground subsidence and climate pattern mapping, including the analysis of vegetation indices (NDVI) and temperature variations, using C, L, and X-band Interferometric Synthetic Aperture Radar (InSAR) satellite data. This involved leveraging Synthetic Aperture Radar (SAR) technology to monitor and assess surface deformations and environmental changes.DOI:10.13140/RG.2.2.31237.68321, conference : https://bit.ly/3ey41Me

• Developed a sentiment analysis model based on BERT using PyTorch, and integrated it with a Flask API for deployment on Google Cloud App Engine via Docker. Additionally, created a project gallery web application utilizing Flask, JavaScript, and Bootstrap for the client-side interface. Developed a 3D mapping web application for urban environments featuring draggable objects, employing BabylonJS and TerriaJS.• Received appreciation & honored with dual track certification: AI and SE from Digital Product School, Unternehmer Technical University of Munich for taking up dual responsibility in unavoidable circumstance.• Trainer at AI-Makerspace, Digital Product School, Unternehmer Technische Universität München workshop on Flask, MapboxJS and MLOps in 2022 under supervision of Dr. Afsaneh Asaei [ https://bit.ly/3xaKVSR ]• Tech Stack: Python, PyTorch, Flask, BabylonJS, TerriaJS, JavaScript, Bootstrap, Docker

• Worked on “6D pose tracking of objects in occlusion or cluttered environments” under Prof Dr. Mehmet R. Dogar in a merit research scholarship from Government of Assam, India. • Contributed in developing novel algorithms for 6D Object Pose Estimation and Tracking for 6 D-o-F robotic arm manipulation planning using RGB-D models.• In-demand research skills: : Pose Estimation, Computer Vision, Segmentation, 3D Geometry and Mathematics.

Contributed in developing novel algorithms for AO peak detection when applied to SCG signal (ECG subspace) in the System Simulation Lab, EEE Department, IIT G with Prof. M.K Bhuyan.• Trained SVM-RBF kernel and achieved an accuracy of 91.30%• Trained custom DNN with Sparse Autoencoder and achieved an accuracy of 91.50%• Built a Flask web-application with GCP to containerize and deploy the Machine Learning model.

Developed CNN architectures for Fine Needle Aspiration Cytology (FNAC) tissue cell samples and made a comparative study for breast cancer classification. Lead Researcher for "Comparative assessment of CNN architectures for breast FNAC tissue samples", Paper (Elsevier): https://doi.org/10.1016/j.tice.2019.02.001 with Kangkana Bora and Prof. Lipi B. Mahanta.• 96.25% accuracy for fine-tuned GoogLeNet-V3 CNN architecture for FNAC dataset.• Comparative assessment with VGG16, VGG19 and ResNet-50 along with their fine tuned versions.