Projects
-
Large Language Model from Scratch(In Progress)
A repo Implementing Large Language Model like ChatGPT from Scratch.
-
Retrieval Augment Generation(RAG): Chat with Documents
Created Interactive interface to chat with documents using a Retrieval-Augmented Generation (RAG) approach. It uses the Attention is All You Need Transformer paper as a primary document, Chroma for vector storage, and OpenAI's language model for chat responses. The application is built using Streamlit to provide an easy-to-use chat interface.
-
Stock Price Prediction Transformer
The project uses a simplified Transformer architecture to forecast stock prices based on historical price data. It includes data loading, preprocessing, model training, prediction, and evaluation phases, with an emphasis on understanding the Transformer's application in time series forecasting.
-
Stock Price Prediction using LSTM
The project uses a simplified lstm architecture to forecast stock prices based on historical price data. It includes data loading, preprocessing, model training, prediction, and evaluation phases, with an emphasis on understanding the Transformer's application in time series forecasting.
-
Eigenface-PCA Facial Recognition System
Implemented eigenface analysis for facial recognition, calculating the mean face and performing PCA from scratch, achieving 85% accuracy rate for face reconstruction with k = 15.
-
Academic Lease: Connecting Students for Lease Transfers
Engineered AcademicLease.com, a social platform for connecting students across 1100+ US universities for lease transfers, including automated CI/CD pipeline for Google Cloud Platform deployment.
-
Leet-Classifier: Automatic Organizing of LeetCode Submissions
Developed 'Leet-Classifier,' a Python-based tool that automates classification and organization of LeetCode submissions by parsing metadata for intelligent sorting, aiding in programming interview preparation.
-
Presidential Speeches Search Engine
Implemented a search engine on a corpus of US presidential inaugural addresses, generating TF-IDF vectors and using cosine similarity for query responses.
-
Python Library for Multi-Layer Neural Networks
Developed a Python library with NumPy for constructing and training neural networks, processing inputs to efficiently output weights, mean error, and predictions.
-
High-Fidelity MNIST Digit Generation with GANs
Developed and trained a Generative Adversarial Network (GAN) on the MNIST dataset, achieving optimal error rates and accurately classifying digits with a discriminator model and generating digit images with a generator model.
-
Enhancing CIFAR-10 Performance with Transfer Learning
Designed and trained a custom CNN model on the CIFAR-10 dataset, assessing improvement through transfer learning with a pre-trained VGG-16 model.
-
Vector Clocks for Event Ordering in Distributed Systems
Created a vector clock mechanism from scratch for event sequencing in distributed environments, ensuring precise coordination across systems.
-
Comprehensive Wallet Management System with Spring Boot
Designed and implemented a comprehensive Spring Boot application featuring a suite of financial transactions, including user registration, login, wallet funding, peer-to-peer money transfers, transaction reversal, and passbook generation. Integrated with MySQL for robust data management and persistence.
