I ended up deploying the system for question extraction + next step generation via a containerized flask app on a local desktop, making it public with ngrok, the deployed system should be live at- https://88ed-207-237-201-41.ngrok.io

• You can invoke the functions (addressed in questions 1 & 2 in the Instruction.md file) via POSTing to the Above URL at either;
  1. /extract_questions
  2. /next_steps
• because of compute constraints and choice, I ended up using GPT-3 for most of the functionality, as a result the system is a bit slow and might take some time to parse through the transcript for the desired input--to--output.

In order to touch on the 3rd item in the deliverables list, I have included two files in the sample_queries dir -- here it shows how to make a request to the API with either cURL or in native python, building the docker container and deploying the API be accomplished if a user has access to a OPENAI_TOKEN and an ngrok account or linux-based machine with docker & public internet port access

• Only include Client/Relevant calls: had a few transcripts where the only folks talking where in house names
  • transcripts like 63a074f3a39c8c8e9c0ebe2d.txt are not really relevant to sales or meaningful client questions, why save them?
• Standardize Names: have a few calls with people like betty white or ben ten -- unless it makes sense to have specific names for some other function or reason we should anonymize these or treat them in the same manner as client 1 or host 1 entries
  • Also had one instance of a weird client name being saved- 'Client 2|Client 2' in the 637cf23dcc7845331578f507.txt file, the function for mapping speakers in transcripts should get looked at