Projects
Selected projects and experiments.
Model Compression for Efficient Artistic Style Transfer
Compresses VGG-19 for neural style transfer using sensitivity-guided fine-grained pruning, achieving 6.64× model compression while preserving stylization quality.
Multi-Agent Agricultural Intelligence with Vision + Retrieval
An AI-powered multi-agent platform delivering farming guidance through weather, market, policy, and disease-analysis workflows with FastAPI, React, and n8n orchestration.
R and Shiny Workflow for Cardiotoxicity Biomarker Insights
Analyzes targeted metabolomics data to identify potential cardiotoxicity biomarkers through statistical analysis, ML-assisted feature selection, and interactive visualization.
Story-Forge: Fine-Tuned LLM Story Generation
2026QLoRA Fine-Tuning and Lightweight LLM Deployment
Fine-tuned Phi-3 Mini (3.8B) using QLoRA with 4-bit NF4 quantization and LoRA adapters, then optimized inference with Q4_K_M GGUF and llama.cpp for lightweight CPU-based deployment.
Rubik's Cube Solver & Visualizer
2025Ilesh Dhall, Prakhar Singh
Interactive 3D Visualization and Algorithmic Cube Solving
Built an interactive Rubik's Cube solver supporting 2×2 to 10x10 cubes, multiple solving algorithms, 3D visualization with Three.js, benchmarking, and step-by-step solution playback.