Text to image convertor
This is a local, offline text-to-image generation application built using Stable Diffusion v1.4 and Streamlit. This tool allows users to generate high-quality images from text descriptions with granular control over inference steps and guidance scale.
AI Ripeness Prediction & Supply Optimization
Built an end-to-end banana ripeness prediction system using PyTorch and a fine-tuned ResNet18 model optimized with ONNX for low-latency inference. Designed a supply chain optimization engine as a Capacitated Vehicle Routing Problem using a high-performance C++ solution with Pybind11 integration.
Dual-Domain Image Restoration (Restormer + FT)
Augmented the Restormer architecture with Fourier Transform layers to process spatial features and filter high-frequency noise. Engineered a custom Cosine Annealing learning-rate scheduler with 50-epoch warm restarts in PyTorch.
Generative AI Pipeline Architect – Stable Diffusion Text-to-Image Web App
Built a real-time generative AI text-to-image application using PyTorch, Hugging Face Diffusers, and Streamlit. Optimized VRAM utilization and latency through dynamic CUDA hardware routing, FP16 processing, and state caching. Created a data provenance pipeline storing multi-image outputs and hyperparameter metadata in JSON for experiment reproducibility.
SMS Spam Detection
Built an LSTM-GRU hybrid deep learning model using TensorFlow and Keras, achieving 97.8% test accuracy and a 0.985 AUC score. Applied tokenization, lemmatization, stopword removal, model tuning, and precision, recall, F1-score, and ROC-AUC evaluation.