projects

Self-Supervised Learning for JEPA

Advisor: Prof.Yann LeCun

Engineered JEPA using Barlow Twins with ResNet-18 Context Encoder and LSTM Cell as Predictor under Markov Assumption. Achieved 3rd Rank in the Competition with a state-of-the-art probing loss of 0.84, which can accurately predict trajectories.

Action-aware REPA for Diffusion based World Models

Advisor: Prof. Saining Xie

Developed Action-aware Representation Alignment (AC-REPA) for action-conditioned diffusion world models by aligining internal representations with Video Foundation Models (VideoMAEv2) to improve rollout dynamics, temporal coherence, reduce artifacts, and enhance planning success in navigation environments.

Multi-Task Fine-Tuning for Code-Mixed Machine Translation

Proposed Multi-Task Fine-Tuning involving first Token-level Language Identification, followed by Machine Translation. Trained Llama 3.2 1B model using QLoRA. Improved BLEU scores of Romanized Hindi by 5.32% compared to simple Supervised Fine-Tuning (Most imbalanced class 0.70% of all tokens).

DevOps and Agile Methodologies Project

Advisor: Prof. John Rofrano

Developed an Admin service to handle Customers of an e-commerce website, following Agile Principles throughout the 5 sprints using Test Driven Development and Behaviour Driven Development. Deployed using Kubernetes, Docker, and Red Hat OpenShift