AI & History (Part 1): The Digital History Stack & The "Atomic Unit"

In Part 1 of this 3-part technical deep dive, we reveal the engineering behind The Venetia Project. How do you build an AI system that can reason over 100-year-old letters without hallucinating?We break down the "Digital History Stack" designed to make thousands of Edwardian primary sources computationally inspectable—moving beyond simple prompting to a grounded, structural approach.In this episode (The Foundation):The "Atomic Unit" Strategy: Why effective RAG starts with breaking archives into single events, not just chunking text.The Stack: Using Gemini 2.5/3.0 for OCR coverage and GPT-4o (Batch Mode) for cost-effective metadata enrichment.Data Engineering: Using AI as a code-generating agent to write custom parsers for messy OCR data.Dual Storage: Why we split the archive into MongoDB (for humans) and Vector Embeddings (for machines).Coming up in the series:Part 2: Querying the Past (Search & Expansion)Part 3: Visualizing the Findings