RAG Text Chunking & Semantic Splitter

{
  "namespace": "default-docs",
  "vectors": [
    {
      "id": "vec_1",
      "values": "[... 1536-dim embedding vector ...]",
      "metadata": {
        "chunk_id": 1,
        "char_count": 174,
        "text": "# Omnikite Architecture Guide\nOmnikite is a 100% private, client-side developer utility platform. Everything runs directly inside your browser without backend server uploads."
      }
    },
    {
      "id": "vec_2",
      "values": "[... 1536-dim embedding vector ...]",
      "metadata": {
        "chunk_id": 2,
        "char_count": 192,
        "text": "## Core Principles\nWe prioritize WebAssembly, native browser APIs, and the Web Crypto API. Because computations happen in local memory, latency is near-zero and data sovereignty is guaranteed."
      }
    },
    {
      "id": "vec_3",
      "values": "[... 1536-dim embedding vector ...]",
      "metadata": {
        "chunk_id": 3,
        "char_count": 270,
        "text": "## Retrieval-Augmented Generation (RAG)\nIn modern AI workflows, large documents must be partitioned into discrete semantic chunks before generating vector embeddings. Proper chunk sizing prevents loss of context while staying within the model's maximum attention window."
      }
    },
    {
      "id": "vec_4",
      "values": "[... 1536-dim embedding vector ...]",
      "metadata": {
        "chunk_id": 4,
        "char_count": 169,
        "text": "### Sliding Overlap\nOverlapping windows ensure that contextual semantics spanning chunk boundaries are preserved in vector databases like Pinecone, Qdrant, and pgvector."
      }
    }
  ]
}