Graph-Based Movie Recommendation System

Designed and implemented an incremental training pipeline to update a Graph RAG-based recommendation system with new user-movie interactions. Optimized FAISS-based vector search for efficient similarity retrieval, reducing redundant computations and improving model update efficiency.

Australia Weather Rain Prediction

Designed and deployed an end-to-end ML pipeline to predict rainfall using Python, Flask, and Scikit-learn,achieving 97% training accuracy on historical weather data.

Containerized the application with Docker and deployed it to GCP Kubernetes Engine (GKE) using YAML manifests, enabling cloud-native scalability.

Prompt To Video

A web application that converts text prompts into animated videos using Manim and Google's Gemini AI.

YouTube Comment Analysis

Developed an end-to-end web application for extracting and analyzing YouTube comments. Implemented sentiment analysis, emotion detection, and toxicity detection using NLP techniques. Conducted time-based comment analysis to identify engagement trends over different periods.

Next Word Prediction Using LSTM RNN

Built an LSTM RNN model for next-word prediction to assist with predictive text. Enhanced model performance through hyperparameter optimization and data augmentation techniques. Achieved 95.67% accuracy, improving text prediction efficiency and fluency.

Deep Learning for Text Generation

Developed and trained character-level LSTM (TensorFlow/Keras, loss: 1.4259) and custom GPT (PyTorch, loss: 1.1561) models on the Shakespeare dataset. Optimized GPT with accumulation of gradients, mixed precision training and top k / top p sampling for the generation of fluent and coherent text generation.

QA Model Library

Created a comprehensive library for question-answering models using various NLP techniques. Implemented multiple approaches including transformer-based models for efficient question answering on diverse datasets.

LangChain Content Summarizer

Created an end-to-end application for summarizing content from various sources, including URLs, PDFs, YouTube videos, and direct text input. Leveraged the LangChain framework to enhance language model processing and efficient content extraction. Designed a user-friendly interface supporting multiple input formats for seamless and accessible summarization.

Anime Recommender System

Anime Recommender System using an MLOps Workflow with a Machine Learning Pipeline, DVC, and Jenkins to deployed on GCP