Visualizing complex character relationships in books has always been challenging for readers. A new tool called Austen aims to solve this problem by generating character relationship diagrams using AI. However, community testing reveals both the promise and limitations of using current AI models for literary analysis.
Austen is an AI-powered Angular application that creates visual relationship maps between book characters using Mermaidjs diagrams. The tool allows users to search any book from Open Library and generate character relationship graphs that can be saved, downloaded, and shared publicly or kept private.
Accuracy Challenges
Users testing the application across various books report mixed results in terms of accuracy. When tested against complex narratives like The Wise Man's Fear by Patrick Rothfuss, the tool recognized many characters but primarily showed their relationships to the protagonist rather than the intricate web between all characters. One user noted that benchmark testing suggests accuracy rates hover around 60% at best.
At best you're looking at 60% or so accuracy
The tool sometimes produces amusing errors, as demonstrated when a user tested Dungeon Crawler Carl and found character relationships that were partially or completely incorrect. For example, the AI incorrectly identified certain allies as enemies and missed key relationship dynamics. Similar issues appeared when users tested classics like One Hundred Years of Solitude, where the AI failed to capture the novel's complex family relationships.
Technical Implementation
Austen leverages multiple technologies including Angular, Analog, TypeScript, and Supabase for its backend infrastructure. For its AI capabilities, the developer initially experimented with Gemini's free tier but found the results disappointing, eventually switching to DeepSeek for better output quality. The system prompt that guides the AI's analysis is publicly available on GitHub, offering transparency into how the tool processes literary information.
Several users suggested potential improvements, such as grounding the AI with a list of actual characters from the story to improve accuracy, or enabling the tool to process EPUB files directly through a Retrieval-Augmented Generation (RAG) system. This would potentially allow for more accurate relationship mapping without requiring the AI to have read the entire book.
 |
|---|
| Austen's GitHub repository showcases the backend technology used in building the character relationship tool |
Practical Applications
Despite accuracy limitations, users have found creative applications for the tool. Some have used it with reference books to get an overview of technical content, with one user noting how database transactions could be visualized as characters with relationships. Others mentioned using similar MermaidJS-based visualization techniques with LLMs for business logic flowcharts, helping to identify corner cases in their designs.
The tool also shows promise for helping readers keep track of characters while reading complex narratives. One user specifically mentioned how this addresses their struggle of having to frequently go back to remember who certain characters are in a story.
Future Potential
Community feedback points to several potential enhancements that could make the tool more valuable. These include handling spoilers by allowing users to specify how far they've read in a book, accounting for changing relationships over time (as enemies become friends or vice versa), and improving accuracy through better AI models or supplementary data.
While Austen may not yet deliver perfect character relationship diagrams, it represents an interesting intersection of AI, literature, and visualization technology. As LLM capabilities continue to improve and with further development based on user feedback, tools like this could eventually become valuable companions for readers tackling complex narratives.
Reference: Austen