Support Chatbot - How Do I Tool

Purpose

Our AI-powered support chatbot is designed to assist FastDraft users by providing instant, accurate answers based solely on our existing Zendesk support articles.

Overview of the Chatbot

• Objective: Improve customer experience by allowing users to self-serve answers to common support questions.
• Data Source: The chatbot indexes and retrieves content only from published Zendesk support articles (no access to tickets, user data, or internal systems).
• Functionality: the chatbot will respond to user queries by surfacing relevant Zendesk article content. No live data, PII, or internal files are used.

AI Safety & Data Protection Controls

To ensure that the chatbot operates securely and within defined data boundaries, several safeguards have been implemented:

Prompt Sanitisation:

• All user inputs are pre-processed to remove potential injection attempts, malicious instructions, or requests for sensitive or out-of-scope data.
• System prompts are tightly controlled and cannot be modified during runtime.

Content Restriction:

• The AI model has access only to the indexed Zendesk articles.
• No connection is made to live systems, databases, or user records.

Azure AI Model Security:

• The underlying model is hosted within Microsoft Azure OpenAI Service, configured with content filtering and abuse monitoring enabled.
• Azure’s responsible AI filters are active to block inappropriate or unsafe responses.
• The model endpoint is locked to specific API keys and IP ranges, ensuring that only the chatbot backend can invoke it.

Data Isolation:

• The entire solution (data, embeddings, and APIs) is deployed in a non-production Azure resource group with network restrictions and private endpoints enabled.
• No cross-environment data transfer occurs between UAT and production.

Logging & Monitoring:

• All API interactions are logged, including prompt inputs and response metadata (no user identifiers).
• Logs are reviewed to monitor model performance, detect anomalies, and ensure compliance with data governance standards.

No Data Retention or Training:

• The chatbot does not store or retain user queries beyond testing logs.
• No conversational data is used to retrain or fine-tune the model.