I built a wiki where you can set the current reading location (book + chapter) and articles get automatically truncated to only show information revealed up to that point. The problem with existing wikis was (for me, at least) that they would spoil complete character arcs already in the first few sentences of an article, basically making it impossible to look up any information before finishing an entire series. For something like the 14-book Wheel of Time series with literally thousands of characters, this is really painful. Especially if you're listening to the audiobooks, spread across months, it's really hard to keep track of all the names and places. There have been only a few attempts of spoiler-safe wikis before, such as Spliki (recently discontinued) and Coppermind (which is really great, but Sanderson-only).
Thulr is still a work in progress. I've only got a single series on there so far, Fourth Wing/Empyrean, which is much less complex than e.g. Wheel of Time. I think the current pipeline won't automatically scale up to those more extreme cases, so I'm trying to only slowly increase the complexity of series that I'm adding. It also currently only supports one clear timeline, so I'm not entirely sure yet how to handle the case where you've got non-obvious reading-orders or auxiliary books (such as having one wiki for all of the Tolkien/Middle-earth books).
My current implementation partially uses LLMs. I built a multi-step pipeline that collects chapter-level information on different entities and then aggregates that into final article candidates. This still needs a ton of manual back-and-forth for quality-control and editing. I want articles to incrementally reveal more information (as opposed to: rewrite the full article on each new time horizon), which makes the amount of text manageable, but can lead to awkward prose at times. E.g., name reveals or character identity collapses (think: Voldemort & Tom Riddle in Chamber of Secrets) can't be spoilered in an article at an early time horizon, but those paragraphs need to stay in the article even at a late time horizon. I'm still trying to find better approaches for this.
Any feedback or ideas on the UX (I'm very new to this) and data processing would be great.
Thulr is still a work in progress. I've only got a single series on there so far, Fourth Wing/Empyrean, which is much less complex than e.g. Wheel of Time. I think the current pipeline won't automatically scale up to those more extreme cases, so I'm trying to only slowly increase the complexity of series that I'm adding. It also currently only supports one clear timeline, so I'm not entirely sure yet how to handle the case where you've got non-obvious reading-orders or auxiliary books (such as having one wiki for all of the Tolkien/Middle-earth books).
My current implementation partially uses LLMs. I built a multi-step pipeline that collects chapter-level information on different entities and then aggregates that into final article candidates. This still needs a ton of manual back-and-forth for quality-control and editing. I want articles to incrementally reveal more information (as opposed to: rewrite the full article on each new time horizon), which makes the amount of text manageable, but can lead to awkward prose at times. E.g., name reveals or character identity collapses (think: Voldemort & Tom Riddle in Chamber of Secrets) can't be spoilered in an article at an early time horizon, but those paragraphs need to stay in the article even at a late time horizon. I'm still trying to find better approaches for this.
Any feedback or ideas on the UX (I'm very new to this) and data processing would be great.