Question regarding the implementation status of specific features

[chaos-03x]

Hi DeepTutor Team,
First of all, thank you for the inspiring paper! I've been following your work on agentic personalized tutoring and found the architecture design particularly impressive.
After diving into the current repository, I noticed some differences between the current implementation and the descriptions in the paper. While the TutorBot infrastructure appears very robust and well-implemented, I would like to clarify a few points regarding the core tutoring engine:

1. Hybrid Personalization: The paper mentions a dual-index structure (Knowledge Graph and dense embeddings) with RRF. However, the current codebase seems to rely primarily on standard LlamaIndex vector retrieval. Are there plans to release the KG-based retrieval and RRF logic?
2. The Trace Forest: I was really interested in the three-level hierarchical memory structure (Session-Plan-Execute). Currently, I see a two-file system (SUMMARY.md and PROFILE.md). Is it scheduled for a future update?
3. Benchmark Replication: Since the current code seems to be a streamlined version, I’m curious if the results reported in the paper (e.g., the 10.8% improvement) were achieved using this specific version or a more feature-complete internal build?

I'm eager to learn from your implementation and potentially contribute to the project. Any guidance or information about the roadmap would be greatly appreciated!