pippin

Here’s a minimal Pippin agent that automates customer support ticket triage. The agent uses a pre-trained NLP model (e.g., `transformers` library) to classify tickets into categories like 'billing', 'technical', or 'feature request', then assigns a priority level (high/medium/low) based on keywords and historical data. It integrates with Zendesk via their API to fetch new tickets and posts updates to a Slack channel for the support team.

**Agent Workflow:**
1. **Fetch Tickets:** The agent polls Zendesk every 5 minutes for new tickets using the Zendesk API (`zendesk-python` library).
2. **Classify Tickets:** It uses a fine-tuned BERT model (loaded via Hugging Face) to classify each ticket’s subject and description into predefined categories. For example, a ticket with 'refund' in the subject is classified as 'billing'.
3. **Assign Priority:** The agent checks historical resolution times for similar tickets (stored in a SQLite database) to assign priority. Tickets with subjects like 'urgent' or 'down' are marked 'high'.
4. **Update Zendesk:** The agent updates the ticket’s tags (e.g., `category:billing`, `priority:high`) and assigns it to the appropriate team via Zendesk’s API.
5. **Notify Slack:** A summary of the classified tickets is posted to a Slack channel (e.g., `#support-triage`) with links to the tickets and suggested next steps.
6. **Learn and Adapt:** After each batch of tickets is processed, the agent logs its predictions and compares them to the actual resolutions (fetched from Zendesk). Misclassifications are used to fine-tune the model weekly.

**Example Output in Slack:**
```
🚀 New tickets triaged (Batch #42):
• Ticket #1234 (Subject: 'Refund request for order 5678')
  - Category: billing
  - Priority: high
  - Assigned to: finance-team
  - Suggested: Contact customer within 2 hours

• Ticket #1235 (Subject: 'Login issues with mobile app')
  - Category: technical
  - Priority: medium
  - Assigned to: dev-team
  - Suggested: Check server logs for auth errors
```

**Python Code Snippet (Pippin Agent):**
```python
from pippin import Agent
import zendesk_api
from transformers import pipeline

class SupportTriageAgent(Agent):
    def __init__(self):
        super().__init__()
        self.classifier = pipeline('text-classification', model='bert-support-triage')
        self.zendesk = zendesk_api.Client(api_token='YOUR_TOKEN')
        self.slack = SlackClient(token='YOUR_TOKEN')

    def fetch_tickets(self):
        return self.zendesk.get_new_tickets()

    def classify_and_prioritize(self, ticket):
        category = self.classifier(ticket['description'])[0]['label']
        priority = 'high' if 'urgent' in ticket['subject'].lower() else 'medium'
        return {'category': category, 'priority': priority}

    def run(self):
        tickets = self.fetch_tickets()
        for ticket in tickets:
            result = self.classify_and_prioritize(ticket)
            self.zendesk.update_ticket(ticket['id'], **result)
            self.slack.post_update(f"Ticket {ticket['id']} classified as {result['category']} (Priority: {result['priority']})")

agent = SupportTriageAgent()
agent.run()
```