Sachidananda Rautray Email & Phone Number

Bhubaneswar, Od, In

Bengaluru, Karnataka, India

• Cleaned and structured pharmacovigilance data in the msafety database using PostgreSQL. Utilized Jaspersoft Studio to create insightful business intelligence reports for data-driven decision making.• Explored GenAI and its associated use cases for Question Answering, summarization, and Named Entity Recognition (NER) using Large Language Models (LLMs), particularly focusing on pharmacovigilance data.• Leveraged various open-source Large Language Models (LLMs) and proprietary Closed… Show more • Cleaned and structured pharmacovigilance data in the msafety database using PostgreSQL. Utilized Jaspersoft Studio to create insightful business intelligence reports for data-driven decision making.• Explored GenAI and its associated use cases for Question Answering, summarization, and Named Entity Recognition (NER) using Large Language Models (LLMs), particularly focusing on pharmacovigilance data.• Leveraged various open-source Large Language Models (LLMs) and proprietary Closed Foundation models to optimize project outcomes. • Developing a large-scale LLM application tailored for the clinical domain involved every stage, from pre-training models from scratch to fine-tuning domain-specific LLMs based on specific requirements. Integration of these models was achieved using frameworks like langchain and llmaindex for tasks including Named Entity Recognition (NER), Question Answering (Q&A), and Summarization.• Implemented advanced techniques including LoRA and QLoRA to fine-tune models such as TinyLlma and Minichat-2-3B, achieving remarkable efficiency by training only 0.70% of the original model parameters using PEFT techniques. Utilized quantization architecture to train models with reduced GPU memory requirements.• Explored techniques like DPO to train the TinyLlma model for machine translation tasks relevant to clinical documents. • Stored document embeddings on vector databases like Pinecone and Chroma for streamlined retrieval and management of extensive document datasets. Show less