Data

Bhagavad Geeta content sources, licensing, and ingestion.

Sources

Primary: gita/gita Repository

• URL: https://github.com/gita/gita
• License: The Unlicense (Public Domain)
• Content: 701 verses across 18 chapters
• Format: JSON with Sanskrit, transliteration, translations

Secondary: VedicScriptures API

• URL: https://github.com/vedicscriptures/bhagavad-gita-api
• License: MIT
• Content: Additional translations and commentaries
• Used for: Enriching verse data with multiple perspectives

Sanskrit Text

The original Bhagavad Geeta verses are ancient texts (~5th century BCE to 2nd century CE), freely usable worldwide without copyright restrictions.

Data Structure

Each verse contains:

{
  "canonical_id": "BG_2_47",
  "chapter": 2,
  "verse": 47,
  "sanskrit_devanagari": "कर्मण्येवाधिकारस्ते...",
  "sanskrit_iast": "karmaṇy-evādhikāras te...",
  "translations": [
    {
      "author": "Swami Sivananda",
      "text": "Your right is to work only..."
    }
  ],
  "commentaries": [
    {
      "author": "Adi Shankaracharya",
      "text": "..."
    }
  ]
}

Ingestion Pipeline

On first run, the backend:

1. Reads verse JSON from data/ directory
2. Validates structure and required fields
3. Inserts into PostgreSQL (verses, translations, commentaries tables)
4. Generates embeddings using sentence-transformers
5. Stores vectors in ChromaDB for semantic search

Manual re-ingestion:

docker compose exec backend python scripts/ingest_data.py --all

Embeddings

• Model: sentence-transformers/all-MiniLM-L6-v2
• Dimensions: 384
• Indexed fields: Sanskrit IAST + English translation
• Storage: ChromaDB with cosine similarity

Attribution

While the Unlicense doesn’t require attribution, we acknowledge:

• gita/gita repository maintainers
• VedicScriptures API contributors
• Traditional commentators (Shankaracharya, Ramanuja, etc.)
• Translators whose work enables access to this wisdom