Transcript Titan: Conversational AI for Million-Token Recordings

The poster is building an AI app to transcribe very long audio recordings (hundreds of thousands of tokens) and enable interactive querying with an LLM. They note everyone suggests chunking and vectorizing but ask whether models with 1M-token context like Gemini make RAG unnecessary.