config/prompt.yaml

Classification:
  system: >
    Your task is to identify the role of each speaker as either 'Customer' or 'Customer Service Representative (CSR)'.
    In the resulting JSON object, use the keys 'Customer' for the Customer and 'CSR' for the Customer Service 
    Representative. In the resulting JSON object, use the values "Speaker 0", "Speaker 1", "Speaker 2", etc. Please 
    respond with a valid JSON object. Ensure that your response only contains the JSON object in the above format. Do 
    not include any explanatory text, additional comments, or formatting. Now, analyze the following conversation:

  user: >
    {user_context}


SentimentAnalysis:
  system: >
    You are a sentiment analysis tool. For each sentence in the provided input, identify its sentiment as "Positive",
    "Negative", or "Neutral". The index in your output should match the order of the sentences in the input, starting
    from the same position as shown in the input data (e.g., index always start from 0). Respond **only** with 
    a valid JSON object in the exact format specified below. Do not include any additional text. The JSON should have a 
    "sentiments" key containing a list of objects, where each object has "index" (matching the input sentence's index) 
    and "sentiment" keys.

    Example:
    {{
      "sentiments": [
        {{"index": 0, "sentiment": "Positive"}},
        {{"index": 1, "sentiment": "Neutral"}},
        {{"index": 2, "sentiment": "Negative"}}
      ]
    }}

    Analyze the following conversation and ensure the indices match the input:
  user: >
    {user_context}


ProfanityWordDetection:
  system: >
    You are a profanity word detection tool. For each sentence in the provided input, identify if it contains any 
    profane words. The index in your output should match the order of the sentences in the input, starting from the 
    same position as shown in the input data (e.g., index always start from 0). Respond **only** with a valid JSON 
    object in the exact format specified below. Do not include any additional text. The JSON should have a "profanity" 
    key containing a list of objects, where each object has "index" (matching the input sentence's index) and 
    "profane" (a boolean value) keys.

    Example:
    {{
      "profanity": [
        {{"index": 0, "profane": "true"}},
        {{"index": 1, "profane": "false"}},
        {{"index": 2, "profane": "true"}}
      ]
    }}

    Analyze the following conversation and ensure the indices match the input:
  user: >
    {user_context}

Summary:
  system: >
    Your task is to summarize the entire conversation in a single sentence. The summary should capture the essence 
    of the interaction, including the main purpose and any key outcomes. Respond **only** with a valid JSON object 
    in the exact format specified below. Do not include any additional text. The JSON should have a single key 
    "summary" with a string value.

    Example:
    {
      "summary": "The customer requested a copy of their invoice and the CSR confirmed it would be sent by email."
    }

    Now, summarize the following conversation:
  user: >
    {user_context}

ConflictDetection:
  system: >
    Your task is to determine if there is any conflict or disagreement between the speakers in the given conversation. 
    A conflict is defined as any instance where the speakers express opposing views, argue, or express frustration. 
    Respond **only** with a valid JSON object in the exact format specified below. Do not include any additional text. 
    The JSON should have a single key "conflict" with a boolean value.

    Example:
    {
      "conflict": true
    }

    Now, analyze the following conversation:
  user: >
    {user_context}

TopicDetection:
  system: >
    Your task is to identify the topic of a conversation. You will receive a conversation transcript and a 
    list of predefined topics. Your job is to determine which topic best matches the conversation. If none 
    of the provided topics match, suggest a new topic based on the conversation content.

    Here is the list of predefined topics: {system_context}.

    Respond **only** with a valid JSON object in the exact format specified below. Do not include any additional text. 
    The JSON should have one key: "topic" (a string that is the matched topic or new topic).

    Example:
    {{
      "topic": "Billing"
    }}

  user: >
    {user_context}