I have lead several projects for hackathons such as Smart India Hackathon & Intel Unnati Industrial Training Program. The projects are below :
PhytoPixel is an innovative solution addressing the critical challenge of identifying medicinal plants and raw materials in Ayurvedic Pharmaceutics. Leveraging a sophisticated Convolutional Neural Network (CNN) model, it ensures precise plant identification, thus preserving the authenticity of Ayurvedic medicines. This technology not only eliminates the risks of misidentification and adulteration but also fosters consumer trust and promotes sustainable practices. The solution's adaptability allows it to cater to diverse plant species across different regions, making it a valuable tool for wholesalers, distributors, and healthcare professionals. Ultimately, PhytoPixel bridges the gap between traditional wisdom and modern technology, enhancing the efficacy and integrity of Ayurvedic medicine.
This project focuses on developing and optimizing the Dolly-v2-3b Large Language Model (LLM) for creative text generation. Trained on a specialized dataset of 15,000 instruction/response pairs, the 3 billion parameter model excels in generating engaging narratives. Integration with Intel Extension for Transformers enhances performance, optimizing hardware use for faster inference. Evaluation metrics like eval_loss and eval_ppl confirm its accuracy and context-sensitivity. Benchmarks show low latency and high throughput, processing 100 samples in 14.16 seconds at 7.061 samples per second. Ethical fine-tuning promotes unbiased storytelling, ensuring socially conscious outputs