Accelerating Drug Discovery with Open Source LLMs: Building a Generative AI-Powered Preclinical Knowledge Engine

Discovering new drugs involves many steps, with preclinical testing being a critical one. This stage includes experiments and studies conducted before testing new drugs on humans. Researchers often struggle to efficiently find the right insights from the vast amount of past studies. Recently, our team has developed an advanced Research Assistant powered by Large Language Models (LLMs). Think of it as a super-smart answer engine designed specifically for researchers. This tool uses a method called Advanced Retrieval Augmented Generation (RAG) to provide precise and relevant answers quickly to the preclinical questions. It is already saving our client's Pharma R&D colleagues a tremendous amount of time!

Unlike traditional search engines that might give you a lot of irrelevant results, our tool understands specific questions related to preclinical testing and finds the most relevant data from past studies and answers questions grounded in the context using LLMs. This significant improvement can speed up the discovery of new drugs, reduce unnecessary experiments, lower costs, and ultimately help bring life-saving medications to patients more quickly. Additionally, this tool assists researchers in writing essential compliance documents much faster, reducing the time required from months to just days.

In this talk, we'll explore how you can build a similar tool using open-source LLMs like LLaMa3 and open-source databases like OpenSearch. We'll discuss the AI techniques involved, customization strategies, and how to take these tools from concept to production. You'll also learn about the positive impact on researchers' efficiency. This session will introduce you with the knowledge to build your own AI-powered knowledge engines, revolutionizing how preclinical research and experimental design are conducted.