We are proud to announce that our colleague Max Schattauer will be a guest speaker at the upcoming Prompt Engineering Conference on November 20th, 2024. Max will give a session about "Improving Retrieval Q&A Contextualization Prompts", using conversation history in general for improving chatbot and Q&A systems' retrieval queries.

In those cases, basing retrieval on the most recent user query alone usually produces less than optimal results. More often, the necessary context is spread across several antecedent interactions. Query contextualization is the process of creating coherent retrieval queries, with relevant context, from message histories.

During this presentation, Max will discuss:

• The top issues that affect retrieval quality: missing context, irrelevant direct answers, and inappropriate follow-up questions.
• How prompt engineering can alleviate these problems by developing robust contextualization models and system prompts using TextGrad.
• Practical examples, deriving from industry-specific cases that allow participants to understand the importance of effective query contextualization within high-quality chatbot solutions.

‍
The session will deliver actionable insights into how to enhance the precision and reliability of AI-driven Q&A systems, showcasing how innovative prompt engineering contributes to superior customer satisfaction.

Further information is available on the official website of the Prompt Engineering Conference.

‍

The results and approaches from the challenge have now been published in the paper "ICML Topological Deep Learning Challenge 2024: Beyond the Graph Domain." This paper delves deeper into the innovative solutions developed during the competition, including our hypergraph-based approach, and explores its applications in fields like medical knowledge graphs, logistics, and business workflows.

The full paper can be found here. For those interested in learning more about Topological Deep Learning, the paper is an excellent resource.

‍