Update README.md

README.md

@@ -1,25 +1,49 @@
 ---
-dataset_info:
-  features:
-  - name: source_link
-    dtype: string
-  - name: title
-    dtype: string
-  - name: chunk_index
-    dtype: int32
-  - name: chunk_text
-    dtype: string
-  - name: embedding
-    list: float32
-  splits:
-  - name: train
-    num_bytes: 93965817
-    num_examples: 4269
-  download_size: 78278965
-  dataset_size: 93965817
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
+  license: apache-2.0
+  language:
+    - en
+  tags:
+    - clickhouse
+    - rag
+    - embeddings
+    - documentation
+  pretty_name: ClickHouse Documentation Embeddings
+  task_categories:
+    - feature-extraction
 ---
+
+  # ClickHouse Documentation Embeddings
+
+  Vector dataset used as the knowledge base for the ClickHouse RAG API. Contains chunked ClickHouse documentation with precomputed embeddings.
+
+  **Not intended for fine-tuning the embedding model.**
+
+  ## Purpose
+
+  This dataset backs the retrieval step of the ClickHouse RAG API. At query time, the API embeds an incoming question using the same model and performs
+  cosine distance search against the `embedding` column in ClickHouse to retrieve relevant documentation chunks.
+
+  ## Schema
+
+  | Column | Type | Description |
+  |---|---|---|
+  | `source_link` | `string` | URL of the source documentation page |
+  | `title` | `string` | Title of the documentation page |
+  | `chunk_index` | `int32` | Position of this chunk within the source page (0-based) |
+  | `chunk_text` | `string` | Text content of the chunk |
+  | `embedding` | `Sequence[float32]` | 4096-dimensional embedding vector |
+
+  ## Embedding Model
+
+  Embeddings were produced by `qwen3_embedding_8b` served via [Text Embeddings Inference (TEI)](https://github.com/huggingface/text-embeddings-inference).
+
+  - **Dimensions:** 4096
+  - **Dtype:** float32
+
+  To query this dataset, embed your query using the same model and configuration. Using a different model will produce incompatible vectors and degrade
+  retrieval quality.
+
+  ## Source
+
+  Sourced from the [ClickHouse documentation](https://clickhouse.com/docs).
+