dataDatasic/historical-culinary-archive-multimodal

Zion Vintage Media: Historical Multimodal Culinary Archive (Sample)

📌 Commercial Licensing & Full Access

This repository contains a 50-item evaluation sample. The full, production-ready dataset contains 10,000+ multimodal pairs (High-Resolution Scans + Structured JSON) of unpublished, offline historical recipes spanning the 1940s to the 1990s.

To purchase a commercial license for the full SQL database and high-resolution TIF/JPG scans for AI/ML training, please contact:

• Email: admin@zionvintagemedia.com
• Website: [zionvintagemedia.com]Vintage store(https://zionvintagemedia.com)
• Showcase: Dataset Explorer

---

📊 Dataset Overview

This dataset provides rare, Alternative Data for training Large Language Models (LLMs) and Vision-Language Models (VLMs). Because these physical, regional church cookbooks have never been published on the public internet, they provide completely novel tokens, historical dialects, archaic measurements, and procedural logic free from standard web-scraping contamination.

Ground Truth Quality

All text has been transcribed and structured from original physical media using high-resolution CZUR archival scanners, allowing for perfect Image-to-Text grounding.

---

🗄️ Data Structure (22 Highly-Enriched Columns)

This sample metadata.csv demonstrates our rigorous data engineering. It includes 22 structured columns:

Recipe Data:

• recipe_id, contributor, contributor_role
• yield_amount, oven_temperature
• ingredients (Serialized JSON)
• instructions & instruction_steps (Serialized JSON)
• special_notes & data_quirks (Captures historical oddities and OCR edge-cases)

Historical Cookbook Context:

• book_title, book_author, publication_year
• location, sales_purpose, historical_notes

File Mapping:

• source_filename, source_filenames (For multimodal image pairing)

---

© 2026 Bot and Dot Retail Management LLC / Zion Vintage Media.