Toxic Comment Detection
A high-precision Deep Learning model achieving 98.01% accuracy in detecting toxic speech. Features an end-to-end pipeline with SMOTE balancing, TF-IDF vectorization, and a Keras Sequential neural network, visualized through an interactive Next.js showcase.
LLM Text Detection
An end-to-end machine learning pipeline distinguishing human vs. AI-generated text with 99.44% accuracy. Built on the DAIGT V2 dataset, this project features a modular architecture, Docker containerization, and automated CI/CD deployment to AWS EC2.
Spam Mail Classification
An end-to-end Deep Learning system deployed on AWS to filter spam with 98% accuracy. Features SMOTE balancing to handle class imbalance, text vectorization, and a custom automated deployment pipeline on EC2.
Chat with Multiple PDFs
An End-to-End Gen AI project that processes 1,000+ pages instantly using Gemini Pro, LangChain, and FAISS. Features a RAG architecture to query multiple PDFs with high precision, deployed via Streamlit.
Building an LLM From Scratch
A research quest to build a GPT-style Transformer from the ground up using PyTorch. Trained on 223k characters of poetry, this project explores self-attention, tokenization, and backpropagation to generate creative text.