openCHS

Appearance

QA Model Documentation ​

1. Model Overview ​

The QA (Quality Assurance) model is a multi-head classifier that is designed to evaluate call center transcripts against a set of predefined quality assurance metrics. It can be used to automate the process of QA and to provide feedback to call center agents.

2. Model Details ​

  • Model Architecture: The model is based on the DistilBERT architecture and has multiple classification heads, one for each QA metric.
  • QA Metrics: The model is trained to evaluate the following QA metrics:
    • Opening
    • Listening
    • Proactiveness
    • Resolution
    • Hold
    • Closing

3. API Endpoints ​

3.1. Evaluate Transcript ​

  • Method: POST
  • Path: /qa/evaluate
  • Description: Evaluates a call center transcript against a set of QA metrics.
  • Request Body: A JSON object with a transcript field.
  • Response: A JSON object with the evaluation results.

4. Request Format ​

The request body for the /qa/evaluate endpoint should be a JSON object with the following field:

json
{
  "transcript": "The call center transcript to be evaluated."
}

5. Response Format ​

The response from the /qa/evaluate endpoint is a JSON object with the following fields:

json
{
  "evaluations": {
    "opening": [
      {
        "submetric": "Use of call opening phrase",
        "prediction": true,
        "score": "✓",
        "probability": 0.95
      }
    ],
    "listening": [
      {
        "submetric": "Caller was not interrupted",
        "prediction": true,
        "score": "✓",
        "probability": 0.92
      }
    ]
  },
  "processing_time": "The time taken to process the request.",
  "model_info": {
    "model_path": "The path to the model.",
    "loaded": "Whether the model is loaded.",
    "load_time": "The time when the model was loaded.",
    "device": "The device on which the model is running.",
    "error": "Any error that occurred during model loading."
  },
  "timestamp": "The timestamp of the request."
}

6. Example Usage ​

bash
curl -X POST \
  -H "Content-Type: application/json" \
  -d '{"transcript": "Agent: Thank you for calling. How can I help you?"}' \
  http://localhost:8123/qa/evaluate

7. Error Handling ​

If the QA model is not loaded, the API will return a 503 Service Unavailable error. If the request is invalid, the API will return a 400 Bad Request error.