AI History Project

AI feels like it was invented yesterday, but its story starts with Turing and survives two "winters" before the LLM era. AI History Project stitches that story together: every concept, person, and organization gets its own page, all linked into a graph that shows how ideas built on each other.

The site is fully static — content is versioned JSON, validated and indexed at build time — and AI is used exactly where it earns its keep: a question-answering assistant grounded in the catalogue, with verifiable citations.

What it does

• Decade-by-decade timeline (1940–2026) with cards for concepts, people, and organizations
• Interactive knowledge graph with 372 nodes and 483 edges
• "Ask AI History": free-form questions answered from the catalogue, with citation chips that highlight graph nodes
• Semantic constellation view: local embeddings projected with seeded UMAP, plus "more like this" suggestions
• Time scrubber animating the field's growth across the decades

Engineering highlights

• The graph is derived at build time from explicit relations, alias-resolved text matching, and a CC0 Wikidata snapshot refreshed monthly by an automated PR
• Q&A is retrieval-grounded with structured outputs, a 300-char input cap, per-IP and daily rate limits, and hides itself gracefully without an API key
• EU AI Act Article 50 transparency: pre-interaction disclosure, data-ai-generated markup, and an X-AI-Generated header
• Static end-to-end: a single Azure Function powers Q&A; everything else is served from the edge