Abhay Kumar

Fine-tuned and deployed custom LLM (LAM) for Agent in conversational context, enabling it to select next step and respond accordingly. This enhancement led to 15% increase in conversation completion ratios. Trained and deployed variety of task-specific Language Models (LLMs) in production, significantly improving system's ability to comprehend user messages. Engineered series of embedding and generative zero-shot models, enabling rapid bot creation and reducing unidentified utterances by 30%. Trained multiple custom language models on domain-specific datasets and got average of 13% improvements on supervised and unsupervised downstream tasks. Developed probability-based spelling correction model, this helped increase accuracy of downstream tasks.

Created semi-supervised Pointwise Learning to Rank model using deep learning algorithms for recruitment industry and Got 30% improvement on baseline model. Developed Pairwise Job recommendation engine that recommends jobs to candidates using deep learning algorithms. This increased candidate's engagement by 25% in tool. Developed internal framework for transformer models in TensorFlow and PyTorch both and pre-trained multiple custom transformer and other transformer-based models on 30 million job descriptions and resume. This reduced experimentation time by 40% in data science team.

Developed sequential classification model using deep learning algorithms for opinion mining, that extracts opinions about chemotherapeutic drugs from medical documents. Developed Information Extraction model using LSTMCRF to extract phrase-based entities among domain-specific documents and this improved precision by 50% on baseline model.

Developed classification model to identify junk part of legal documents. Developed crawling pipeline responsible for crawling legal websites and parsing HTML documents and storing relevant structured information in database.

Co-Founded e-commerce marketplace with some new ideas like where customers can negotiate price of products with sellers.