coldbydefault-portfolio
    Preparing search index...

    Function searchKnowledge

    • Hybrid search over the retrieval index: cosine similarity plus Postgres full-text search, merged by reciprocal rank fusion. Embeddings alone miss one name in a long list ("does he know Drizzle?" scores below the cut-off against his skills), which keyword matching catches.

      Blog chunks are returned only while their post is published, so drafts (which are embedded too) never reach Reem. No language filter: embeddings match across languages, and each result says which language it is in. The corpus is a few hundred rows, so both rankings are computed in full.

      Parameters

      • query: string

      Returns Promise<KnowledgeSearchResult[]>