Featured Projects

Explore projects showcasing my expertise in Machine Learning, AI models, and Exploratory Data Analysis (EDA).

Anime Nexus

A sophisticated recommendation system blending collaborative and content-based filtering to deliver personalized anime suggestions, built with a robust MLOps pipeline for scalability and deployment.

Frameworks: PyTorch, Flask, Jenkins, Docker, Google Kubernetes Engine, DVC, Comet ML, GCP

GitHub View Deployed App
Multilingual Assistive Model for Visually Impaired Users

Supports visually impaired users by integrating image captioning, translation, and text-to-speech in multiple Indic languages.

Frameworks: Pytorch, Blip's model, Transformers, IndicTrans2, gTTS API, HuggingFace Spaces

GitHub View Deployed App
UvU Net: An Ensemble Generator for Image Colorization

UvU-Net is an advanced ensemble model for sketch-to-photo colorization, leveraging a dual U-Net architecture with residual connections.

Frameworks: PyTorch, CycleGAN, U-Net, Bahdanau Attention, Deep Learning

GitHub
Hotel Cancellation Prediction

A production-grade MLOps pipeline using LightGBM to predict hotel booking cancellations, with automated CI/CD, Docker containerization, and serverless deployment on Google Cloud Run.

Frameworks: LightGBM, Flask, Jenkins, Docker, Google Cloud Platform, MLflow

GitHub View Deployed App
SPARQL Query Generation
Natural Language to SPARQL Query Generation

Designed a system that converts natural language questions into SPARQL queries for retrieving answers from Wikidata. It utilizes an NMT model trained on the QALD-9 dataset.

Frameworks: PyTorch, Seq2Seq, Bi-LSTM, Attention Mechanisms, SPARQL, NLP

GitHub
Membership Inference Attack on GANs

Implemented a black-box Membership Inference Attack on GANs to determine if a data point was used in training, using a shadow model and membership classifier.

Frameworks: PyTorch, NumPy, scikit-learn, matplotlib

GitHub
Diffusion-based Adversarial Purification Pipeline
Diffusion-based Adversarial Purification over Latent Embeddings

Developed a defense against adversarial attacks using latent diffusion, achieving 43.4% robust accuracy on PGD attacks (ε=16/255) on ImageNet.

Frameworks: PyTorch, torchvision, NumPy, Pillow

GitHub