From Search Box to Semantic Brain: Building AI-Native Search in TYPO3

What if TYPO3 could search by meaning and answer questions from its own content instead of just matching words? This talk shows how ML-based semantic search can turn TYPO3 records into a searchable knowledge layer. Using the extension smart_search as a technical case study, we follow the path from CMS content to embedding vectors, from cosine similarity to intent-based retrieval, and from retrieved records to grounded AI answers with RAG. We will also look at how local models can be integrated for privacy-aware and self-hosted AI workflows, and which practical ML decisions matter most: representation quality, chunking, thresholds, stale embeddings, and model-specific vector spaces.