Completed Projects

These are only some of my favorite projects; many more can be found in my GitHub Repositories.

Multi-Modal AI Music Assistant    

    AI, Machine Learning, Recommendation Systems 2023

A personalized music recommendation system powered by AI that combines facial emotion detection, contextual awareness, and music analysis. The system detects user emotions from facial images, considers environmental factors (weather, time, location), and uses advanced Matern kernel similarity metrics for music matching. It connects with Spotify for seamless playback and adapts recommendations based on user feedback.

                 

Llama-3.1-8B-Sarcasm: Human-Like Conversational AI  

    Natural Language Processing, LLM Fine-tuning 2024

A fine-tuned language model based on Meta’s Llama-3.1-8B-Instruct that generates natural, human-like responses in conversations. The model excels at generating contextually appropriate replies with sarcasm, humor, and casual tones. Trained using AdaLoRA (Adaptive Low-Rank Adaptation) for parameter efficiency and improved generalization, it can process both single-turn exchanges and multi-turn conversations up to 7 turns of context. Available in both full model and quantized GGUF formats for efficient deployment.

                 

Swiggy Food Delivery Analysis & Prediction    

    Data Science, Predictive Analytics 2023

A comprehensive analysis of Swiggy food delivery data with predictive modeling to forecast delivery times, customer satisfaction, and order patterns. The project combines exploratory data analysis with machine learning techniques to identify key factors affecting delivery performance and customer experience, providing actionable insights for optimization.

                 

NumPy Transformers: English-Hindi Translation  

    Deep Learning, NLP, Transformers 2024

A complete implementation of transformer architecture from scratch using NumPy/CuPy for English to Hindi translation. The project features a full encoder-decoder architecture with RoPE (Rotary Position Embedding), multiple optimizer options (SGD, Momentum, RMSProp, Adam, Nadam) with Noam scheduler, and multi-head attention mechanisms with visualization support. The implementation includes batched processing with masking, checkpoint saving, and comprehensive performance metrics tracking.

             

Multi-Agent Portfolio Generator  

    AI Agents, LLM Applications 2025

An advanced multi-agent system that transforms user documents into a fully functional portfolio website. The system utilizes specialized agents working in coordination to analyze user documents, extract key information, design an appropriate portfolio structure, generate clean and functional code, and provide clear implementation instructions. This agentic approach enables completely automated portfolio creation tailored to the user’s experience and achievements.