Abstract: ChatGPT is the fastest-growing user application in history. Still, this application only has access to information it saw during training and sometimes produces false information, called hallucinations.
In this talk, we will show you how to bring your data to LLMs and how to evaluate LLMs for your use case using open-source technology. We will give practical insights into retrieval augmented generation, the technique used to bring your company data to LLMs. You will learn how to evaluate the effect of prompt engineering, how to improve the retrieval part in retrieval augmented generation, and how to reduce model hallucinations. Those techniques can be applied to any domain and even to a variety of natural languages. We will showcase examples from industry clients who use the open-source framework Haystack to connect data and users with LLMs for building applications running in production.
Bio: Timo Möller is Co-Founder of deepset and Head of Solution Engineering. He works closely together with deepset's clients to bring modern NLP into production. He is an open-source fan and a passionate NLP engineer. Currently, he works on retrieval augmented generation, auto-generating training data, and ways to detect hallucinations.