Vector databases arrived as a category, acquired enormous funding, and produced a great deal of documentation that assumes you already understand why you need one. Weaviate Academy is a genuine attempt to fix that, and it is better than most of what its competitors have built. The academy is organised as courses rather than as a documentation tree, which is the right decision and a surprisingly rare one. Key Concepts and Architecture runs about an hour and covers what a vector database is doing and why the architecture looks the way it does.
Hands-on Weaviate with Python is three hours and builds a working application. A Gentle Deep Dive into AI Models covers embeddings and the model side. Your First AI App covers search and RAG. An architectural decisions guide covers the choices you have to make when you set a collection up.
There is an agents unit listed as coming soon, which in vendor documentation terms means anywhere between next month and never. The AI models unit is the one I would send people to first, including people who have already picked a different vector database entirely. Understanding what an embedding actually represents, why cosine similarity is the usual choice, what happens to your results when you change embedding model, why chunking strategy dominates retrieval quality more than any database setting, all of that is portable knowledge and it is taught here clearly and without much product placement. Most people building RAG systems are weak on exactly this and it shows in their results.
The hands-on Python course is well constructed. It builds a movie search and recommendation API rather than running disconnected snippets, which means you finish with a mental model of how the pieces fit rather than a folder of examples. The free cloud sandbox removes the setup friction that kills a lot of database tutorials before they start, and being able to get a working instance in two minutes matters more for learning than it sounds like it should. The architectural decisions material deserves credit for existing at all.
Choosing a distance metric, deciding on your index configuration, working out whether you need multi-tenancy, deciding how to handle updates, these are the decisions that are painful to change later and that most tutorials skip entirely because they are boring compared to the demo. Weaviate covers them. Now the limitation that no vendor academy escapes, and it is worth being blunt about. Every question here has a Weaviate shaped answer.
Nowhere in this material will you find a serious discussion of when you should not use a vector database. That discussion matters, because a great many teams reach for one when a Postgres extension would have done, or when their corpus is small enough that a brute force scan in memory is faster and simpler, or when their actual problem is a keyword search problem wearing a fashionable hat. Weaviate is a good product and it is also frequently the wrong tool, and the academy has no incentive to say so. Hybrid search is covered from the angle of how to configure it rather than the angle of when dense retrieval is failing you and why.
In practice hybrid search rescues a lot of RAG systems precisely because pure vector search is bad at exact terms, product codes, names and acronyms. That is a limitation of the approach, and it is presented here as a feature to enable rather than a weakness to understand. Evaluation is the real gap. The single hardest part of building retrieval systems is knowing whether yours is any good.
Building an evaluation set, measuring recall at k, understanding where retrieval fails versus where generation fails, running an experiment when you change chunk size, none of this is taught in any depth. It is also the thing that separates a demo from a system, and its absence is the main reason I would not treat this academy as sufficient on its own. The certification has been in development for long enough that I would not plan around it. That is a minor complaint, since a Weaviate certificate would carry limited weight anywhere regardless.
Client language coverage is uneven. Python is thoroughly served, everything else considerably less so, and if you are building in TypeScript you will spend more time in the reference documentation and less in the courses. The reference docs themselves are decent, reasonably current, and reorganised recently, and they are what you will actually live in once you are past the courses. My four out of five is for a vendor education programme that genuinely teaches rather than demonstrates, with material on embeddings that stands on its own merits, held back by the structural honesty problem that applies to all vendor training and by the missing evaluation content that anyone shipping retrieval into production will need to find elsewhere.