This project spans Oct 2023 - May 2024. The focus is to build a multi-modal and multi-media AI system with a voice/text UI to extract relevant data from financial statements, create summaries, perform Exploratory Data Analysis (EDA), create charts/graphs, and analyze differences across financial documents.

The goal of our study is to develop "BurgerBot," a framework combining GPT-4 and a segmentation model to guide a robotic arm in assembling a plastic burger toy. The objective is to understand real-time human-robot conversational interactions, focusing on the robot's adaptability to diverse instructions.

The goal is to build an end-to-end NLP system involving collecting our own data and training a model to identify specific entities such as method names, task names, dataset names, metric names and their values, hyperparameter names and their values within scientific publications from recent NLP conferences (ACL, EMNLP, and NAACL).

Develop a minimalist version of BERT (Bidirectional Encoder Representations from Transformers), implementing some important components of the BERT model (self attention, layers, model, optimizer, and classifier) to perform sentence classification on sst dataset and cfimdb dataset.

This project leverages the Multi-Scale Feature Fusion Network (MSFFN), a CNN-based model that extracts and fuses features at different scales for nuanced classification. It achieves 80.64% accuracy, outperforming transformer-based baselines with just 1.1 million parameters, making it lightweight and cost-effective. The ensemble variant, MSFFN-Ensemble, further enhances performance, achieving 86.82% accuracy, showcasing the power of domain-specific and generator-specific learning.