Inspectable AI Reasoning
Research Question
How can learned models work with structured knowledge and explicit constraints in tasks where the reasoning process matters as well as the final output?
Research Approach
Our research interests include neurosymbolic methods, structured representations, rule-based checks, and evaluation of intermediate reasoning artifacts. These approaches may help identify unsupported steps and expose where a system needs stronger evidence or human review.
Evaluation Focus
We examine consistency, traceability, failure cases, and the relationship between intermediate representations and final answers. Inspectable artifacts are useful only when their relationship to actual system behavior is tested, rather than assumed.
Technologies and Methods
Neurosymbolic methods, structured knowledge, explicit constraints, and reasoning evaluation.