Tried the demo. The concept makes sense for telecom R&D, where information is often buried in large technical documents. I’m curious how you handle grounding when the source material is very dense or math-heavy, and whether the system can return precise citations instead of broad summaries. If you are open to sharing more about your architecture or data approach, that would be interesting to read.