Financial AI can now rank documents without training — but simpler prompts beat complex systems

Researchers built a system that uses large language models to search financial documents without requiring any model training. In practice, this means financial analysts can deploy AI search tools immediately using open-source models, without spending months on data labeling and model tuning.

Financial firms have been waiting for AI that works out of the box on their documents. This paper shows it's possible — and more importantly, shows which tricks actually work and which are theater. The finding that simple prompt engineering beats complicated multi-agent systems tells practitioners what to actually build, not what sounds impressive. That's rare enough to matter.