Technical Skills: Machine Learning, Computer Vision, Remote Sensing, Data Engineering, Distributed Systems.

Tech Stack: Python, PyTorch, Pandas, NumPy, Linux, CLI, Git, Selenium, Docker.

• B.Tech(Computer Science) | IIIT Delhi(Aug 2019- May 2023)
• High School(STEM) | Prince International School

Data Scientist @ Jio Platforms Limited (June 2023 - Present)

• Built an Energy Usage forecasting pipeline and utilized it for Distributed Energy Management optimization using constraint programming.
• Utilized IoT tower data to develop a battery backup estimation model, enhancing energy usage efficiency in Jio Telecom Towers.

Research Assistant @ Vision Lab, IIITD (August 2022 - May 2023)

• Curated a remote sensing dataset of 3 satellites, comprising 200k+ images from 350+ agriculture fields.
• Built a crop monitoring system using remote sensing data and deep computer vision.
• Leveraged multi-modality in satellite data for multi-sensor fusion, resulting in better performance.

Research Assistant @ TavLab Lab, IIITD (August 2022 - May 2023)

• Leveraged language models to understand COVID-19 gene and quantify mutation in it.
• Used web scraping to automate the data collection process and using multiprocessing parallelized data ingestion pipeline, leading to 10x increment in team’s productivity.

WACV’24(ORAL) | Depanshu Sani, Sandeep Mahato, Sourabh Saini, Harsh Kumar Agarwal, Charu Chandra Devshali, Saket Anand, Gaurav Arora and Thiagarajan Jayaraman, . "SICKLE: A Multi-Sensor Satellite Imagery Dataset Annotated with Multiple Key Cropping Parameters."

Vision Transformer | Multi-Modality | Adversarial Training

• Used ViT to encode images and GPT-2 as text decoder for image captioning.
• Tested models on Adversarial examples to assess robustness of the model.

Self Supervised Learning | Adversarial Training | Computer Vision

• Improved SOTA model convergence by 4x through introducing Minkowski distance measure.
• Used low curvature activation function to handle over-fitting issue of pre-trained model.

Machine Learning | Sentiment Analysis

• Employed diverse language models to generate meaningful vector representations, leveraging them to construct a sentiment classification model..
• By using non-heuristic based approach we achieved similar performance as reference paper with 20x less parameters.