Jessica
Jessica — AI assistant
Ticket analysis, difference analysis and learning rules for improving AI responses
Access
Jessica is available in the Studio of the Cusmato dashboard, under the Jessica tab. She appears as a chat interface next to the ticket view.
What Jessica can do
Ticket analysis
- Retrieve tickets — Get ticket data from Redis (outbox or completed)
- Batch analysis — Analyse multiple tickets at once
- Quality analysis — Analyse the quality of generated responses with user feedback
Difference analysis
Jessica can analyse the difference between:
- The AI-suggested response
- The actually sent response (after corrections by the agent)
This helps recognise patterns and improve the AI system.
Learning rules
Jessica can save learning rules that improve future AI responses:
- Categories — tone, structure, content, formatting
- Rule — Clear instruction in natural language
- Examples — Optional: ticket IDs where the rule applies
These rules are used by the auto-improve system to optimise the system prompt.
Processing feedback
When agents give feedback on AI responses:
- Feedback is stored in the pending feedback queue
- Jessica can analyse these corrections
- Auto-improve synthesises approved corrections into an optimised prompt
Tools
Jessica uses the following tools (via function calling):
| Tool | Description |
|---|---|
get_ticket_data | Retrieves ticket data (caseId, status: outbox/completed) |
get_all_outbox_tickets | Retrieves all outbox tickets for review |
get_tickets_with_feedback | Retrieves tickets that have received feedback |
analyze_response_quality | Analyses response quality with feedback |
save_learning_rule | Saves a learning rule (rule, category, examples) |
get_learning_rules | Retrieves saved learning rules (optionally filtered by category) |
Auto-improve
The auto-improve system (separate API endpoint) analyses approved corrections and generates an optimised system prompt. It:
- Categorises corrections by theme (tone, information, structure, personalisation)
- Prioritises by impact (high, medium, low)
- Integrates changes into the existing prompt
- Validates coherence and completeness
See AI model - Auto-improve for more details.
Practical tips
- Ask for analysis— "Analyse the outbox tickets" or "Show tickets with feedback"
- Save learning rules— After recognising a pattern: "Save: for complaints always show empathy for the frustration"
- Difference analysis— "Compare the AI response with the sent response for ticket X"