Developed and trained a Convolutional Neural Network (CNN) from scratch to classify handwritten digits from the MNIST dataset. The model achieved an outstanding 99.3% accuracy on the 10,000-image test set.

This project was built using Python, TensorFlow, and Keras , and the model’s robust performance across all digit classes is confirmed by the detailed classification report and low test loss.

This project addresses the critical business problem of customer churn for a telecom company. The primary goal was to analyze customer data to identify key factors leading to churn and to build a predictive model to flag at-risk customers.

After performing extensive data manipulation and visualization, four different classification models were developed and evaluated. As the comparison chart shows, the Multiple Logistic Regression model delivered the best performance, achieving a 79.7% accuracy in predicting customer churn.

This repository is a comprehensive collection of my practical work from my intensive Machine Learning and AI training. It serves as a personal log of implementing fundamental algorithms and concepts from scratch, covering a wide range of topics from classical machine learning to deep learning.

Key topics covered include:

• Supervised Learning: Linear Regression, Logistic Regression, Decision Trees, SVM.
• Unsupervised Learning: K-Means Clustering.
• Deep Learning: Building and training Neural Networks with TensorFlow & Keras.

This project is a custom-built conversational AI assistant with an interactive web interface created using Streamlit. It provides a direct and responsive chat experience, leveraging the power of OpenAI’s GPT models to engage in natural and coherent conversations.

The application showcases the ability to successfully integrate a powerful large language model with a user-friendly front-end, creating a seamless and real-time chatbot experience.