---

name: document-modeling description: "Model multimodal data with DocArray BaseDoc schemas, predefined documents, DocList, DocVec, nested fields, and typed tensors." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

DocArray document modeling

Use this sub-skill when a task needs to define, validate, or batch multimodal DocArray data models.

Route here for

• Designing BaseDoc schemas for text, image, audio, video, embeddings, metadata, and nested multimodal payloads.
• Choosing predefined document classes such as TextDoc and ImageDoc versus custom BaseDoc subclasses.
• Creating dynamic document classes with create_doc() or create_doc_from_dict().
• Choosing DocList versus DocVec for streaming, reranking, shuffling, or contiguous ML batch processing.
• Modeling nested documents, nested DocList fields, optional fields, and shape-checked NdArray tensor fields.
• Handling optional Torch, TensorFlow, or JAX tensor type boundaries without claiming those optional backends are verified.

Route elsewhere

• Serialization protocols, protobuf/JSON/bytes/CSV/DataFrame round-trips, file stores, S3, or FastAPI response payloads: use sibling serialization-storage.
• Vector indexes, nearest-neighbor search, document indexes, or external vector database backends: use sibling vector-indexing.

Read these bundled references

1. Start with API reference for verified imports, constructor signatures, core class behavior, and the modeling decision table.
2. Use Workflows for task recipes and copyable examples.
3. Use Troubleshooting when validation, Pydantic config, DocList, DocVec, NumPy, or optional tensor backends fail.

Safe bundled helper

Run schema_smoke.py in an environment where DocArray is installed to confirm the main schema, dynamic document, DocList, typed tensor, and optional DocVec paths:

Run these commands from this sub-skill root (the directory containing this SKILL.md):

python scripts/schema_smoke.py --help
python scripts/schema_smoke.py
python scripts/schema_smoke.py --skip-docvec

The verified minimum scope is CPU with DocArray base plus proto, pandas, and web extras. Optional services and tensor frameworks are documented as boundaries, not verified defaults.