Boardside

The Flow Problem

Modern board games are complex. A single rules dispute can derail an entire session—players dig through 50-page PDFs, search unreliable forum threads, or argue interpretations until someone gives up. I built Boardside to solve this as a board game hobbyist who was tired of losing momentum mid-game. The goal was simple: get a definitive, citation-backed answer in seconds without leaving the table. But there was a secondary motivation: testing whether an AI agent could reliably navigate "legal" language—rules, FAQs, and errata—where precision matters and context is everything.

The RAG Challenge

The hardest part wasn't building a RAG pipeline—it was chunking. Rules documents are dense and context-dependent. Chunk too small, and you lose the surrounding logic needed for a ruling. Chunk too large, and semantic search becomes imprecise. I built a system using LangGraph and FastAPI that chunks rulebooks into semantic sections stored in Supabase, balancing granularity with context. The app supports text-to-speech and speech-to-text so you can ask questions hands-free. Currently indexing 16 games, with an automated ingestion pipeline ready to scale.

Early Traction

I posted about Boardside in a single Board Game Geek thread and got 300+ users in two days. The demand validated the pain point: players want fast, trustworthy answers without breaking immersion. The next phase is expanding the library and refining the chunking strategy to handle edge cases in even more complex games.