Cognizant - Generative AI Extern

93.3% Accuracy

Achieved high accuracy in dog breed classification using VGG model

30% Faster

Reduced training time while maintaining model performance using PEFT

Multiple Models

Implemented and compared RESNET, VGG, and ALEXNET performance

As a Generative AI Extern at Cognizant, I focused on developing and implementing sophisticated AI solutions. My first major project involved creating a dog classifier using pre-trained models including RESNET, VGG, and ALEXNET. Through rigorous testing and optimization, we achieved a remarkable 93.3% accuracy rate with the VGG model.

A significant achievement was the development of a Tweet evaluator using advanced natural language processing techniques. By implementing PEFT (Parameter-Efficient Fine-Tuning) with the Lora Config method on a roberta-based model, we successfully improved emoji classification accuracy while reducing training time by 30%.

The project utilized the comprehensive Hugging Face ecosystem, including Transformers, Datasets, Evaluate, PEFT, and Scikit-learn. This experience enhanced my expertise in efficient model adaptation and fine-tuning for specific tasks.

Python Deep Learning Hugging Face PEFT Model Fine-tuning Data Analysis