Optimal Multimodal Chunking & Timestamp Granularity for Audio/Video + Long Docs in Lumora KB?

While evaluating Lumora KB on 60-minute internal training recordings alongside companion SOP PDFs, we noticed: - Fixed 30-second transcript windows often split causal explanations; - Semantic chapter chunks (3–5 minutes) yield much higher retrieval precision, but operators still want second-level playback jump links. What is the recommended two-tier indexing setup? Does Lumora support parent-chunk semantic recall with child word-level timestamp anchoring?