Self-supervised Video Frame Prediction and Segmentation

Part of Deep Learning Course - Prof. Yann Lecun, Prof. Alfredo Canzianni Deployed a U-Net architecture for semantic segmentation on the objects of interest within the final predicted frame; created a self-supervised framework to predict next frames with a SimVP (CNN-based encoder-decoder) model. Achieved top 20 percentile[Jaccard index 0.25] in DL course competition of predicting and segmenting future video frames.

Semantic Role Labeling using BiLSTM

Part of NLP Course - Prof. Adam Meyers Developed a deep learning model using BiLSTM for Argument Prediction (object and subject) of a sentence on Nombank Dataset; based on dependency mappings and POS tags; used Tensorflow library for training on cloud GPUs. Our model outperformed the baseline model and achieved a precision of 0.88 and F1-score of 0.82.

Efficient Communication Portal for Mentors and Students

Java Servlets, JSP, MySQL

Created an engaging communication platform for mentors and students with features, including leave applications, meeting scheduling and performance tracking; along with admin side features like allotment of students to mentors. Implemented advanced features like relocating/updating students assigned to mentors and tracking their history.

Smart Activity Management System for Professors

Node.js, MongoDB, Vue.js

Developed a platform to automate manual work of professors and admins; designed features like keeping track of monthly activities conducted and attended such as workshops, seminars; teaching schedule management, along with admin side features like monitoring professors’ records, adding or removing professors from the system and generating excel reports

Android Application for Cafeteria Management

Java, Firebase, Android Studio

Implemented features like user authentication, providing daily menus to students, pre-ordering meals, maintaining order history, integrated payments and credits systems. Utilized firebase as database for admin, students, menus and credits.