Making Large Legislative Records Accessible with AI

The Challenge

• Ontario Legislature has massive amounts of transcript data
• 20M+ tokens of parliamentary proceedings
• Too large to fit in LLM context windows
• Need efficient way to find and load relevant content

Our Solution: Smart Context Management

We use two different LLM models: one that helps to retrieve relevant content, and one that answers the question. We leverage Gemini's context caching to cut down on costs and compute time. See docs.md for more details.