Demo video and live system for the CAIS'26 demo Scaling Expert Feedback with Reflective Edit Propagation in Compositional Knowledge Bases.
Walkthrough video:
The video walks through a typical correction-and-propagation cycle: an expert opens a low-confidence test description, edits it inline, and reviews the agent's stepwise plan (intent inference → symbol update → search → batch revision) before committing the propagated changes.
Link: https://raid-demo.azurewebsites.net/
The live demo is loaded with the RxTerms pharmaceutical dataset (~20k entries) used in our quantitative evaluation. It runs the same reflection-propagation pipeline described in the paper. Note: this is a development environment.
Guo, J., Li, X., Ono, J. P., He, W., & Ren, L. (2026). Scaling Expert Feedback with Reflective Edit Propagation
in Compositional Knowledge Bases. In ACM Conference on AI and Agentic Systems (CAIS '26).