Roadmap (1:17:56)
| Area | Item | Note |
|---|---|---|
| Enterprise readiness | RBAC support | Called "boring stuff but necessary stuff" (1:17:56) |
| Hybrid search at scale | Block-Max WAND | Replaces plain WAND pruning (1:17:56) |
| Multi-vector retrieval | ColBERT / ColPali support | These use many vectors per token, requiring MaxSim aggregation plus efficient pruning (1:17:56) |
| Efficiency | Locally adaptive quantization | The item the speaker is "super excited about" — shrink the memory footprint without degrading performance. He compared quantization techniques in a guest lecture at Brown earlier that year (1:17:56, 1:18:51) |
| Generative | Generative feedback loops | The self-updating database concept from the opening (1:18:51) |
| Product | Recommendations as a service | Goal: useful recommendations, not merely the closest matches (1:18:51, 1:19:14) |
- Weaviate recruits for summer internships and full-time positions (
1:19:14).
Closing endorsement
- The lecturer's opinion: among all the vector databases out there, Weaviate has the best blog and the best documentation, because they are open about what they actually do — and open source (
1:19:38). - Contrast drawn: other vector databases can be "a bit coy" about their internals and use terms that don't line up with what everyone else calls the same thing (
1:19:52).