Building a Voice-Controlled Graph Assistant With Neo4j, LiveKit, and OpenAI

Build a voice assistant that lets users speak natural language questions to a Neo4j Aura book graph and hear answers spoken back — all from a Jupyter notebook. The stack combines LiveKit Agents for WebRTC and voice pipeline, OpenAI Whisper and TTS for speech, and neo4j-graphrag's Text2CypherRetriever for query generation. Key engineering challenges include resolving a package version conflict, crafting schema prompts with type cast hints, and using Jupyter's %%writefile to avoid pickling failures.