| name | docling |
| description | Expert guidance on document processing with Docling and audio transcription with Whisper. |
Docling & Ingestion Expert Skill
This skill provides patterns for processing diverse document formats and audio files.
📝 Document Processing
- Multi-format support: Use
docling for converting PDF, DOCX, PPTX, XLSX, HTML, and Markdown to a clean Markdown representation.
- HybridChunker: Always use Docling's
HybridChunker to preserve document structure (headings, lists, tables).
- Token Awareness: Configure chunker with
max_tokens=512 to align with embedding models (e.g., text-embedding-3-small).
🔊 Audio Transcription
- Whisper ASR: Use
docling's integration with Whisper Turbo for transcribing audio files (.mp3, .wav, .m4a, .flac).
- Path Handling: Pass
pathlib.Path objects to DocumentConverter for audio files.
🏗️ Pipeline Architecture
- Ingestion Pipeline: Centralize logic in
src/ingestion/ingest.py.
- Metadata Extraction: Extract YAML frontmatter and standard document properties.
- Batching: Use batch embedding generation and
insert_many for chunks to optimize performance.
