Project Title: Building and Evaluating Robust Foundation Models for Human Genomics• Built four language models using encoder techniques of different sizes on protein sequences ranging from 500M up to 2.5B parameters • Pre-trained on three datasets including a collection of diverse human genomes (3,202), and genomes from 850 unique species• Performed Probing techniques using the Nucleotide Transformer models which out performs 12 out of 18 genomic downstream task • Achieved an AUC score on non-coding variant predictions using the DNA-based transformer of 0.73 on eQTLs and 0.71 on meQTLs

Project Title: Artificial Intelligence and TensorFlow for High Speed Video Analysis• Used Python tkinter library to create a GUI for visualization of High speed video of Gas Metal Arc Welding (GMAW) in cine format• Stored the videos of GMAW processes for globular and spray transfer modes as a 4D array (time, width, height, channels) into a .npz file • Applied Transfer learning using a Unet Model for video segmentation of GMAW welding process to calculate physical properties• Merge the model to the GUI, with options for the physical properties like, velocity and acceleration of droplets, arc length

• Garnered 223 (48% YoY increase) Foreign Training offers with projects from leading universities like Standford, Imperial College etc• Organized On Campus round of Hult Prize, 2023 mentored by 5 judges and 3 speakers which witnessed the participation of 40+ Teams

Helped Office of International Relations in organizing semester exchanges, semester away and hosting international delegations.

Project Title: Application of Machine Learning Techniques in Aircraft Rescheduling Problem• Automated column generation technique to create dataset comprising various disruption scenarios of 50 aircrafts on a 7-day schedule • Generated 58 features using domain knowledge, applied feature engineering and backward selection technique to identify aircrafts• Applied various models including Decision Tree, Random forest classifier, Xgboost for subset generation of possible carrier aircraft• Achieved 0.78 recall with basic models, employing SMOTE with CatBoost and hyperparameter tuning, the best result reached 0.92