coldbydefault-portfolio
    Preparing search index...

    Variable ragConfigConst

    ragConfig: {
        chunkOverlap: 150;
        chunkSize: 1000;
        minScore: 0.25;
        rrfK: 60;
        siteUrl: "https://coldbydefault.com";
        topK: 5;
    } = ...

    Reem's retrieval index (lib/rag). The embedding model comes from the EMBEDDING_MODEL env var; with it unset, Reem has no search tool and indexing is skipped. The model's dimension is fixed in vector(3072) on KnowledgeChunk.embedding (prisma/schema.prisma): changing the model means a schema change and a full re-index.

    Type Declaration

    • ReadonlychunkOverlap: 150
    • ReadonlychunkSize: 1000

      Target chunk length in characters, and how much of the previous chunk repeats.

    • ReadonlyminScore: 0.25

      Cosine similarity below this is dropped unless the chunk also matches a query keyword. With text-embedding-3-large, unrelated text scores ~0.15–0.2.

    • ReadonlyrrfK: 60

      Reciprocal rank fusion constant: higher flattens the gap between ranks.

    • ReadonlysiteUrl: "https://coldbydefault.com"
    • ReadonlytopK: 5