PRISM: Prompt-Refined In-Context System Modelling for Financial Retrieval

PRISM is a training-free framework for financial information retrieval that combines prompt engineering and in-context learning to rank documents and chunks. It systematically tests three components—prompting, ICL, and multi-agent coordination—finding that a simpler prompt-refined system is most reliable for production use.

Open the full episode

More from AI Radio