May 4, 2025
I did a live stream this morning about voice AI with the amazing @RayFernando1337. We started off with a demo of the open source Word Wrangler game. Word Wrangler's code demonstrates a couple of patterns that are useful for complex, multi-LLM, realtime voice applications.
When we started the demo, the first phrase was "neural network." Which we did not plan. There are ~600 words in the game's word list. 🤔
Word Wrangler is driven by two Gemini Multimodal Live API pipelines:
1. one pipeline implements a character who is guessing the phrases that you're describing, and,
2. the other implements a judge who is independently checking the answers, keeping score, and telling you (but not the other LLM) each phrase.
Complete source code: https://t.co/gg9rMS0oxU
The live stream with Ray: https://t.co/IdJxbQc62K