TritonGPT Chatbot Widget

Overview

The TritonGPT Chatbot Widget offers a powerful and cost-effective way to provide AI-powered support and information retrieval on UC San Diego websites hosted on the campus CMS. Built on the TritonGPT platform and integrated with the UC San Diego Assistant, the widget allows users to interact through natural language directly on a webpage. This capability is part of UC San Diego’s broader strategy to leverage generative AI to improve administrative efficiency, enhance the digital experience, and make university services more accessible to all.

Purpose

This widget serves as a new modality for engaging with TritonGPT. It enables departments to embed an interactive AI assistant into their campus CMS websites, offering users immediate and contextual support. This functionality is particularly useful for public-facing websites, where visitors may have questions but are not authenticated UC San Diego users.

Functionality

• Drop-in interface: The widget is lightweight and easy to embed on any UC San Diego-managed website.
• Natural language understanding: Users can type questions in everyday language and receive relevant responses based on TritonGPT’s institutional knowledge.
• Multi-language support: Users can ask questions in their native languages and receive relevant responses based on TritonGPT’s institutional knowledge.
• UC San Diego Assistant integration: The assistant draws from a curated knowledge base that includes over 60 UC San Diego websites, ensuring accurate and context-aware answers.
• No authentication required: The widget works seamlessly for both internal and external users, making it ideal for admissions, student services, HR, research centers, and other outward-facing sites.
• Campus-aligned alternative: This provides a campus-supported alternative to commercial chatbot products, which are often expensive, complex to maintain, and lack contextual understanding of UC San Diego-specific content.

How We Keep Chatbot Conversations Secure

To ensure a safe, secure, and reliable experience, all messages sent to the AI Chatbot widget pass through an automated, multi-layered security system before being processed by the AI. This system acts as a digital checkpoint, filtering for inappropriate content and preventing misuse of the service.

Here’s how it works:

• Authentication: First, your chat session is given a temporary, anonymous security token. This initial step helps prevent service abuse.

• Jailbreak and Misuse Detection: Next, your message is scanned by a specialized filter designed to detect attempts to “jailbreak” or trick the AI into bypassing its safety rules and core instructions.

• Harmful Content Filtering: Finally, the message is analyzed by a second, comprehensive filter that checks for a wide range of harmful or illicit content across 14 different categories, including hate speech, threats, and defamation.

Only messages that successfully pass all security checks are sent to the AI to generate a response. If a message is flagged at any step, it is immediately blocked, and a generic error message is returned. This entire process happens almost instantly to protect users and maintain the integrity of the AI service.

Blink Use Case

On Blink, the chatbot widget enhances usability by offering intelligent, immediate responses to questions that arise as users explore policies, processes, or services. It helps reduce navigation friction and supports quicker access to relevant content.

Get Started

If your department is currently paying for a third-party chatbot service or considering one, we encourage you to reach out to explore whether the TritonGPT widget could serve as a more integrated, cost-effective alternative.