[MVP] Feynman session

[KenjiPcx]

Context

This is right after the session configuration phase. Here we attempt to explain our chosen topic to the agent

Requirements

UI

• Transcribe speech using Azure Speech Service (currently used) React Speech Recognition or use Whisper for better transcripts
• Somehow send transcripts periodically to agent, make it as natural as possible, could be debounced based (wait 10s until user stops talking), an extra idea I had was using a gpt3.5 instance as a transcripts manager to autocorrect bad transcripts and decide whether a user is still speaking or not to send to a backend function
  • (Nicolo) Convoscope debounces on the frontend and backend, which I think is confusing and they do it only for faster processing for entities

Backend

• Backend function could be api/process_speech or similar and it would return a response as a student
  • Currently, it responds with the student's response, an expert's opinion criticizing the user's explanation and an emotion (I thought this could show the learner's expression in real time)
• Backend function gets thread_id through session_id and interacts with OpenAI's assistants api
• Save transcripts and response to the database, we can display line by line critique in the post session screen

Child Issues

• [x] #14
• [x] #15
• [x] #30
• [x] #16
• [x] #51