I am a skilled Machine Learning professional with a strong background in Natural Language Processing (NLP), Speech Recognition and Computer Vision. My experience includes training and deploying machine learning models in production environments for customised applications. Natural Language Processing:- Extensive experience in working with language models. Pre-trained BERT-base model from scratch on 10 billion hinglish sentences (insurance domain) using Masked Language Modeling (MLM) objective. - The pre-trained model was fine-tuned for various downstream tasks including Question-Answering, Named Entity Recognition (NER), Text Classification, Sentence Embedding and Semantic Search Application. Similar experiments were performed with T5-small and T5-base models for comparison.- Hands-on experience in fine-tuning and continued pre-training of large language models (Mistral, Llama3, etc.) using QLoRA for custom applications and domain adaptations. The fine-tuned model was deployed in production using vLLM for faster inference.Speech Recognition:- Pre-trained the Wav2Vec2 base model from scratch on 10k hours of telephonic conversations, including modifications in the Feature Encoder of the model to process the raw audio sampled at 8 kHz. The pre-trained model was fine-tuned for speech recognition and language identification tasks.- Trained SepFormer model for speech enhancement and audio source separation.Computer Vision:- Expertise in training models for image classification, object detection and image segmentation across different scenarios. - Trained the FaceNet model for facial recognition using Triplet Loss.- Trained EfficientNetV2-small and MobileNetV3 (edge device deployment purpose) models for weeds and crop disease image classification at Purdue University.- Trained the YOLO-v3 model for detecting different types of defects in sheet metal at Suzuki Motor Gujarat.My diverse skill set and practical experience in Machine Learning enable me to contribute effectively to various innovative projects.