Projects
Machine Learning Library from Scratch
Implemented core ML algorithms from scratch, enhancing my understanding of data science foundations.
- Built Linear & Logistic Regression, KNN, Decision Trees, K-Means, and N-layer Neural Networks
- Used only NumPy for implementation, no high-level ML libraries
- Tested and documented on Google Colab for reproducibility
SmartMail - RAG-Powered Email Assistant
Automated email responses using Retrieval-Augmented Generation (RAG).
- Built using Python and LLM integration
- Extracts context from email chains and responds smartly
- Ideal for smart inbox automation
MNIST Digit Classifier from scratch
Implemented a digit classification system from scratch to achieve a 97.2% accuracy on test set.
- Modularized each components into separate files to ensure reusability.
- Model architecture with optimizers are plug and play.
- The NeuralNetwork class dynamically chains all layers and performs full forward and backward propagation from sratch.
Image Caption Generator
Developed a deep-learning model and deployed it as a public API that generates natural-language captions from images.
- Built an image captioning pipeline featuring MobileNet-based encoders with a custom Bahdanau attention layer, improving contextual relevance.
- Implemented both greedy and beam search decoding, offering a spectrum of speed vs. caption quality trade-offs.
- Exposed model capabilities through a Flask-based REST API, enabling image upload and caption retrieval through dynamic decoding mode selection.