TWI Map uses an AI-assisted pipeline to process all 807 chapters (~12 million words) of The Wandering Inn web serial, identify named geographical references with Claude, and render 600+ filtered locations as an interactive Leaflet map with chapter-based spoiler controls. A single binary embeds all assets through go:embed.
What This Project Shows
This project shows the source, output format, review steps, operating cost, and released result for one recreational corpus. It is not customer past performance or a production authorization.
Features
- Spoiler-free chapter slider: show locations mentioned up to your current chapter
- Multi-format navigation for web, audiobook, and ebook readers
- Clickable relationship lines between connected locations
- Ghost provenance lines showing extraction source
- Searchable sidebar with location details
- Keyboard navigation, ARIA labels, screen-reader announcements, and WCAG AA text contrast
- Single binary deployment: all assets embedded
Pipeline
The four-stage extraction pipeline:
- Scrape: Download chapters from the web serial
- Extract: Use Claude (Sonnet 4) to identify and extract geographical references with structured output
- Aggregate: Deduplicate, normalize, and cross-reference locations across all chapters
- Serve: Present an interactive Leaflet map with chapter-based filtering
By the Numbers
“AI-generated code” describes the implementation method. A person specified the result, reviewed behavior and accessibility, validated the output, and remains responsible for the release.
Tech Stack
Go 1.25+, Cobra CLI, DuckDB for storage, Leaflet.js for mapping, Claude API for extraction. Requires an Anthropic API key for the extraction step.