Music Genre Classification

2024

Developed a PyTorch-based audio classification model using transfer learning on ResNet50, processing multi-resolution spectrograms of song snippets to distinguish between progressive rock and non-progressive rock.

Combined deep learning with logistic regression on the final layer, achieving 0.75 AUC on the test set.

Created pipelines including audio preprocessing, snippet-based classification, and song-level prediction aggregation.

Computer Vision Tasks

2023

Implemented diverse computer vision models using PyTorch, including ResNet for image classification and U-Net for semantic segmentation.

Engineered a Neural Radiance Fields (NeRF) model for novel view synthesis and a depth estimation system for 3D scene understanding from 2D images.

Developed YOLO (You Only Look Once) object detection model from scratch using PyTorch, demonstrating proficiency in real-time object recognition algorithms.

Dirrect Democracy App

2022

Developed a React Native mobile application with Firebase backend, enabling citizens to propose, discuss, and vote on local bills anonymously or publicly.

Implemented key features including bill proposal, voting mechanisms, discussion forums, and customizable user profiles to enhance civic participation.

Integrated recommendation, search, and sorting algorithms in Python to improve content discovery and overall user experience.

Crime Detector App

2021

Engineered a deep learning model using TensorFlow and Keras, integrating VGG16 for transfer learning and LSTM for sequence analysis, achieving 87% accuracy in violence detection.

Implemented a robust data pipeline, processing video frames using OpenCV and automating data retrieval and preprocessing for model training.

Designed and deployed a Django-based web application for live camera monitoring and real-time violence detection, enhancing security surveillance capabilities.

HappyLlama - Sentiment Analyser

2020

Developed a Flask-based web application using TensorFlow and scikit-learn, achieving 91% accuracy in sentiment classification of sentences.

Implemented an end-to-end ML pipeline with automated data retrieval using Selenium, integrating collection, preprocessing, training, and deployment.

Created a user-friendly interface for real-time sentiment analysis, optimizing performance for efficient web-based predictions.