• Currently working on building GPU based Implementation for Identification and Classification of Infrared Image for UAVs for DRDO.• Enhanced the performance of the segmentation model by improving its Attention layer, resulting in a higher IOU 94% from 88% for Salient Infrared Small Target Detection.• Leveraged ILNET, FreeU and YOLO models to solve complex real-world problems in the defence sector by fine-tuning model architecture to enhance detection performance in challenging infrared environments.

• Containerized Orfeo Toolbox (OTB) using Docker, integrating it with Python for seamless remote sensing operations. Resolved complex dependencies and optimized environment variables in scientific computing setups.• Worked on optimizing the Segmentation Model to extract flood areas from SAR (Sentinel-1) images.• Developed a detailed Map documenting major floods occurring in 2023 by leveraging data extracted from Twitter.• Contributed to the development of a full-stack website serving as a SAR data inventory generator, leveraging Planet API for researchers' convenience.• Contributed to a project focused on developing a web application for tourists, enhancing their exploration of new destinations through geolocation-based audio guidance.

• Developed and implemented a Deep Q-Neural Network (DQN) model to optimize speed management in autonomous vehicles.• Analysed the impact of varying traffic densities, various vehicle speeds, and number of controlled vehicles on the model's effectiveness and did extensive hyperparameter tuning and exploration fraction to enhance model performance.• Implemented experience replay technique to improve training stability and efficiency.• Achieved significant improvements in episode lengths, rewards, and overall training stability.• Presented findings in a research poster at the Google CSR event, showing potential for real-world application in enhancing the safety and efficiency of autonomous vehicles.

• Developed Deep learning pipeline to extract tweets about flood related locations in India, Pakistan region and send their location on real time to rescue teams.• To accomplish this, employed open-source software named snscraper to extract tweets. Subsequently, utilized the state-of-the-art Bert-Large-NER model, which surpassed the Bert-Base-NER and the NLTK library-based model's performance to accurately recognize key flooded locations in India.• The extracted coordinates of flooded locations were then mapped using the Map box API, facilitating efficient coordination with rescue teams with accurate coordinates.

• Currently working with Mr. Dolon Banerjee on the project “Groundwater Recharge prediction using Machine Learning for Punjab Region”• Conducted extensive literature review on the topic of groundwater recharge, machine learning techniques and geospatial data.• Developed machine learning model for predicting groundwater recharge using LSTM, MLP, and Random Forest with r2_score of 0.87 and MSE of 0.15. This involved analyzing large amounts of government data related to environmental and climatic factors such as rainfall, temperature, soil, LULC, and slope, Latitude, Longitude ,District.• Multiple machine learning libraries were used to develop and train the model. This included data preprocessing, feature engineering, model selection and hyperparameter tuning.

I am gaining experience with the Google Machine Learning frameworks, platforms, and tools that are required by professional developers.Here I learned about the beauty of Deep learning and fine-tuning them and implemented them using TensorFlow and Keras. • Built and trained various Neural Network architectures such as Convolutional Neural Networks, Recurrent Neural networks, LSTMs, and Transformers.• Applied TensorFlow for Multiclass classification, improving accuracy on MNIST using Convolutions.• Participated in Kaggle Competitions organized by Google• Got TensorFlow developer Certification and Deep Learning Professional Certification.

I did my Research Internship under the premier SPARK program at the Mehta Family School of Data Science and Artificial Intelligence at IIT Roorkee with an acceptance rate of 0.5% out of 15000+ applicants under Prof. Alok BhardwajOn the topic of Identification of Flood Extents using Synthetic Aperture Radar images.Work Achievements :• Successfully segmented and identified flooded areas from Urban flooded SAR (Synthetic Aperture Radar) images taken from Sentinel-2.• Used different deep learning pre-trained models for Transfer Learning including LeNet-5, Google LeNet, Inception V3, EfficientNetB7, ResNet-50, MobileNetV2, and VGG-19, ZFNet for the same, and learned to implement the YOLO algorithm version 3.• I did segmentation, morphology, binarization, Contouring, and other operations of OpenCV to achieve my results.• Presented poster titled “Identification of Flood Extents using Synthetic Aperture Radar images”.• Currently working on publishing our work in a renowned journal.

*Built a full-fledged working website for a startup company*Led a team of 5 undergraduate students in founding a startup from scratch in the span of 2 months(as it was an assigned project)*Got in Top 6 Startup Ideas of IndiaI had a great time while learning lots of new skills in internwell and throughout the bumpy road of learning, there was huge support from Ayush Thakur(Co-founder of Internwell) and Sachin Korla (Co-founder of Internwell ).