Autopreso turns speech into live presentations

This is less a polished product than a proof that the realtime voice-plus-agent stack is now good enough to make live deck generation feel practical. The rough edges matter, but the workflow is genuinely useful for anyone who wants a presentation to build itself while they talk.

• –The staging-canvas idea is the right guardrail: seed the model with context first, then let live speech drive incremental edits
• –Buffered transcript chunks plus tool use is a sensible architecture for avoiding jittery, one-token-at-a-time nonsense
• –The local Moonshine path lowers the barrier for tinkering, but the OpenAI path will likely be the more usable one for now
• –This is most compelling for demos, teaching, and rapid visual note-taking, not for fully autonomous slide authoring yet