Inspiration
From a board game enthusiast's point of view, nothing is worse than repeatedly pausing game night to clarify or look up rules within an (often giant) rulebook tome. Games take longer, people lose interest, vibes are killed.
What it does
Magister answers any board game question and lets you easily import, organize, and search through your collection.
How we built it
We built an automatic ingest that crawls the web looking for board games and rulebooks. We then parse the data into structured, source-linked evidence for the AI to easily sift through.
Under the hood, Magister uses Cloudflare Workers, D1, and R2; Convex for published game data; TanStack Start for the web; and SwiftUI for iOS.
Challenges we ran into
- scouring the web for official rulebooks (because sometimes publishers don't have them on their website)
- getting box art for games
- coming up with a search/retrieval strategy that allows the AI to get full context of a question and not just surface the first thing it finds
Accomplishments that we're proud of
We built an experience that puts official rulings up front without hiding the evidence. This is an inherently boring workflow that we intentionally hide from the user with some beautiful game visuals and collection-related animations.
What we learned
Simple search isn't good enough for rulebooks. Often rules are interconnected in different ways and making sure the agent gets enough context to come up with the correct ruling is challenging.
What's next for Magister
We want to keep expanding the game library, improve coverage of tricky questions, and make the experience of collecting games more delightful. The goal is simple: less time searching, more time playing.
Built With
- codex
- convex
- swift
- tanstack
Log in or sign up for Devpost to join the conversation.