New ReSS framework combines symbolic decision trees with LLMs to create interpretable AI models that reason through tabular data while maintaining human-understandable explanations.

ReSS bridges symbolic and neural reasoning by using decision-tree models to extract decision paths that guide LLM reasoning

The framework generates grounded natural-language explanations that strictly adhere to underlying decision logic for transparency

Addresses key challenges in high-stakes domains like healthcare and finance requiring both accuracy and verifiable, interpretable predictions

Fine-tunes pretrained LLMs into specialized tabular reasoning models using high-quality datasets curated from symbolic scaffolds

Tackles dual problems of scalable data curation and reasoning consistency in domain-specific applications