Oracle AI Database 26ai can run AI models next to your data, except the token-level kind that Named Entity Recognition needs. This article covers the need, the limitation, the experiment that removed it with my own Arabic NER model, and a comparison with OCI Language and the Private AI Services Container.
SQL distance metrics confirm that CLIP’s shared vector space is geometrically organized. A Gradio UI confirms whether the results are meaningful to a human reviewer. This post builds that prototype and uses it to surface CLIP’s real constraints on a car damage dataset: class confusion in the dent category, complete accuracy inversion on French-language glass shatter queries, and a search engine that returns results for every input regardless of domain fit.
Oracle 26ai can act as a compact environment for multimodal visual similarity experiments. This post shows how to use real HuggingFace image/text data, Oracle’s pre-built CLIP ViT-B/32 ONNX models, VECTOR columns, and SQL-based similarity search to build and validate an image similarity pipeline inside the database.
Before experimenting with CLIP-based image and text similarity in Oracle 26ai, you need data that is real enough to produce meaningful results. Oracle’s documentation examples are toy-scale; production claims data isn’t ready for a local POC. HuggingFace is the answer. This post shows exactly how to import tahaman/DamageCarDataset into Oracle 26ai and wire up the table structure that the entire CLIP experiment series runs on.
A practical local setup for validating Oracle AI code, testing vector-search workflows, and building a repeatable 26ai experimentation environment. Setting up a local Oracle Database 26ai Free environment is one of the most practical ways to start experimenting seriously with the new AI-related capabilities of the database without waiting for access to a larger managed environment. While working on chapters for my new co-authored Packt book on Oracle AI, I needed a repeatable setup to validate code, test ideas, and refine examples before they made their way into the manuscript. A local Docker-based lab turned out to be the simplest and most effective answer. It gave me a controlled environment for testing AI Vector Search, ONNX model loading, SQL-based embeddings, and early database-side RAG experiments without introducing unnecessary operational overhead. ...
Welcome to assoudi.blog. This is where I publish practical, technically grounded writing on Oracle AI. My goal is simple: useful Oracle + AI content for practitioners, architects, and technical leaders.