This model performs a semantic search of windsurf text resources based on a users query. It accepts a json user query and returns the nearest resources with the url of origination. The motivation is to provide a source of truth that can be relied upon when learning windsurf techniques. The returned json could be used as a source of truth for an LLM prompt to reduce the risk of hallucinations.
The search works by the resources and query being embedded into a vector of 768 dimensions that capture meaning through context. The facebook FAISS algorithm compares resources and queries by generating the dot product of the embeddings. It sees how aligned the embedding vectors are for resource and query. The higher the dot product the more align. For example, completley orthangonal would be reducing each other, whereas fully aligned would be reinforcing each other.