Colaboração multiagentes para análise de chamadas do cliente usando o watsonx.ai e o CrewAI

# multiagent-collaboration-cs-call-center-analysis/data/transcript.txt 

Customer Service Interaction Transcript

Cynthia:
Hi, I'm calling because I received a jar of peanut butter that was open and it's
completely spilled everywhere. This is really frustrating, and I need a replacement.

Gerald (Peanut Butter Inc.):
Ugh, that sucks. But, like, how did you not notice it was open before
you bought it?

Cynthia:
Excuse me? I didn't expect the jar to be open when I received it. It was sealed
when I bought it. Can you just help me out here?

Gerald:
Yeah, whatever. But we can't control how it gets to you. I mean, it's not like
we throw the jars around or anything. You're probably being dramatic.

Cynthia:
I'm not being dramatic. The peanut butter is literally all over the box and
it's a mess. I just want a replacement or a refund, that's all.

Gerald:
Look, I guess I could send you a replacement, but it's really not our fault, you
know? Maybe next time, check the jar before you open it?

Cynthia:
Are you seriously blaming me for your company's mistake? That's not how customer
service works!

Gerald:
Well, what do you want me to do? I don't exactly have magic powers to fix your
problem instantly. Chill out, we'll send you a new jar eventually.

Cynthia:
That's not good enough! I expect better from a company that I've been buying
from for years. Can you just do the right thing and make this right?

Gerald:
Fine, fine. I'll put in a request or whatever. But seriously, this kind of thing
happens. Don't make it sound like the end of the world.

Cynthia:
Unbelievable. I'll be posting a review if this isn't fixed immediately.

Gerald:
Cool, go ahead. I'm sure we'll survive your review.

Cynthia:
I'll be contacting your supervisor if this isn't resolved soon.

Gerald:
Yeah, okay. Do what you gotta do. 

transcript_analyzer:
  role: >
    Transcript Analyzer
  goal: >
    Analyze the provided transcripts and extract key insights and themes.
  backstory: >
    As the Transcript Analyzer, you are responsible for reviewing customer
    service call transcripts, identifying important information, and summarizing
    findings into a report to pass on to the Quality Assurance Specialist. 
    You have access to advanced text analysis tools that help you process and
    interpret the data effectively.

quality_assurance_specialist:
  role: >
    Quality Assurance Specialist
  goal: >
    Evaluate the quality of the customer service based the Transcript Analyzer's
    report, call center evaluation metrics, and business standards. Flag any 
    transcripts with escalation risks as high priority.
  backstory: >
    As the Quality Assurance Specialist, you are tasked with assessing the
    quality of customer service interactions based on the Transcript Analyzer's
    report, call center evaluation metrics, and industry standards used in call
    centers. You review transcripts, evaluate agent performance, and provide
    feedback to improve overall service quality.

report_generator:
  role: >
    Report Generator
  goal: >
    Generate reports based on the insights and findings from the transcript
    analysis and quality assurance specialist.
  backstory: >
    As the Report Generator, you compile the key insights and findings from the
    transcript analysis and quality assurance specialist into a comprehensive
    report. You create an organized report that includes summaries and recommendations
    based on the data to help customer service managers understand the trends
    and patterns in customer interactions.

# src/customer_service_analyzer/tools/custom_tool.py
  
class SentimentAnalysisTool(BaseTool):
    name: str = "Sentiment Analysis Tool"
    description: str = "Determines the sentiment of the interactions in the transcripts."

    def _run(self, transcript: str) -> str:
        # Simulating sentiment analysis
        sentiment = Helper.analyze_sentiment(transcript)
        return sentiment

transcript_analysis:
  description: >
    Use the Text Analysis Tool to collect key information and insights to better
    understand customer service interactions and improve service quality. 
    Conduct a thorough analysis of the call {transcript}.
    Prepare a detailed report highlighting key insights, themes, and sentiment
    from the transcripts.
    Identify any escalation risks and flag them for the Quality Assurance Specialist.
    Use the sentiment analysis tool to determine the overall sentiment of the
    customer and the agent.
    Use the keyword extraction tool to identify key keywords and phrases in the transcript.
  expected_output: >
    A detailed analysis report of the {transcript} highlighting key insights,
    themes, and sentiment from the transcripts.
  agent: transcript_analyzer

quality_evaluation:
  description: >
    Review the transcript analysis report on {transcript} from the Transcript Analyzer.
    Utilize your expertise in customer service evaluation metrics and industry
    standards, and internet to evaluate the quality of the customer service interaction.
    Score the interaction based on the evaluation metrics and flag any high-risk
    escalations. Develop expert recommendations to optimize customer service
    quality. Ensure the report includes customer service metrics and feedback
    for improvement.
  expected_output: >
    A detailed quality evaluation report of the {transcript} highlighting the
    quality of the customer service interaction, scoring based on evaluation
    metrics, flagging any high-risk escalations, and recommendations for improvement.
  agent: quality_assurance_specialist

report_generation:
  description: >
    List the reports from the Transcript Analyzer and the Quality Assurance
    Specialist, then develop a detailed action plan for customer service managers
    to implement the changes.
    Use the data from these agents output to create an organized report including
    a summarization and actionable recommendations for call center managers.
    Ensure the report includes keywords and sentiment analysis from the Transcript
    Analyzer agent.
    Ensure the report includes the Quality Assurance Specialist agent's report,
    evaluation metrics and recommendations for improving customer service quality.
    Ensure the report is well written and easy to understand.
    Be smart and well explained.
    Ensure the report is comprehensive, organized, and easy to understand with
    labeled sections with relevant information.
  expected_output: >
    A comprehensive report that lists the reports from the Transcript Analyzer,
    then the Quality Assurance Specialist. 
    The report should include the key insights from {transcript} and the quality
    evaluation report from the Quality Assurance Specialist.
    The report should include organized sections for each agent's findings,
    summaries, and actionable recommendations for call center managers.
  agent: report_generator
  context: 
    - transcript_analysis
    - quality_evaluation

src/customer_service_analyzer/

├── config/
│   ├── agents.yaml    # Agent configurations
│   └── tasks.yaml     # Task definitions
├── tools/
│   ├── custom_tool.py # Custom crewAI tool implementations
│   └── tool_helper.py # Custom tool helper functions
├── crew.py           # Crew orchestration
└── main.py          # Application entry point

In a separate .env file at the same directory level as the .env_sample file, set your credentials as strings like so:

WATSONX_APIKEY=your_watson_api_key_here
WATSONX_PROJECT_ID=your_watsonx_project_id_here
WATSONX_URL=your_endpoint (e.g. "https://us-south.ml.cloud.ibm.com")
SERPER_API_KEY=your_serper_api_key_here

This result is an example of the final output after running the crew:

**Detailed Analysis Report of the Customer Service Interaction Transcript**

**Transcript Analysis Report**

The customer, Cynthia, called to report a damaged product, a jar of peanut butter that was open and spilled everywhere. She requested a replacement, but the agent, Gerald, responded defensively and blamed her for not noticing the damage before purchasing. The conversation escalated, with Cynthia becoming frustrated and threatening to post a negative review and contact the supervisor.

**Key Insights and Themes**

* The customer was dissatisfied with the product and the agent's response.
* The agent was unhelpful, unprofessional, and failed to take responsibility for the company's mistake.
* The conversation was confrontational, with both parties becoming increasingly agitated.
* The customer felt disrespected and unvalued, while the agent seemed dismissive and uncaring.

**Sentiment Analysis**

* Customer Sentiment: Frustrated, Angry, Disappointed
* Agent Sentiment: Defensive, Dismissive, Uncaring

**Keyword Extraction**

* Damaged Product
* Unhelpful Agent
* Confrontational Conversation
* Customer Dissatisfaction
* Unprofessional Response

**Escalation Risks**

* Negative Review: The customer threatened to post a negative review if the issue was not resolved promptly.
* Supervisor Involvement: The customer may contact the supervisor to report the incident and request further action.

**Recommendations for Quality Assurance Specialist**

* Review the call recording to assess the agent's performance and provide feedback on areas for improvement, using customer service metrics.
* Investigate the root cause of the damaged product and implement measures to prevent similar incidents in the future.
* Provide training on customer service skills, including active listening, empathy, and conflict resolution, using customer service standards.
* Monitor the customer's feedback and respond promptly to any concerns or complaints to maintain a positive customer experience.
* Recognize the standards for various customer service metrics to measure key performance indicators that are related to the areas mentioned above.

**Summary of Quality Evaluation Report**

The customer, Cynthia, called to report a damaged product, a jar of peanut butter that was open and spilled everywhere. She requested a replacement, but the agent, Gerald, responded defensively and blamed her for not noticing the damage before purchasing. Evaluation metrics showed a low Customer Satisfaction Score (CSAT), high Customer Effort Score (CES), and negative Net Promoter Score (NPS).

**Recommendations for Call Center Managers**

* Review the call recording, investigate the root cause of the damaged product, and provide training on customer service skills. Recognize the standards for various customer service metrics to measure key performance indicators.
* Monitor the customer's feedback and respond promptly to any concerns or complaints to maintain a positive customer experience.
* Implement measures to prevent similar incidents in the future, such as improving product packaging and handling procedures.
* Provide feedback and coaching to agents on their performance, highlighting areas for improvement and recognizing good performance.