Orchestrator Design

[Beilinson]

1. Orchestrator reads from the **audio queue **, which the RT STT writes to
2. The orchestrator manages the state of what is currently happening (idle listening, active listening, executing Tasks / LLM, STT)
3. During Idle Listening, if any sentence written into the audio.txt file doesn't contain the start keyword WILSON,
   then the sentence gets yoinked.
4. Once WILSON is said, state transititions to active listening.
5. In Active Listening, the orchestrator is conducting two possible options:
   • Searching for a possible Task to execute
   • Feeding the input to the LLM process.
6. After the [SILENCE] sentence, the orchestrator transitions to the Execute Stage, after which the result can get fed to the STT.
7. After the STT stage, even if an active Task is running, we return to the Idle Listening Stage