Dormibot: Resident Assistant Support AI

1. Objectives and Topic of Dormibot

Dormibot is an AI-powered assistant designed to support Resident Assistants (RAs) in their duties by providing quick access to policies, procedures, and contact information. The chatbot aims to:

• Provide instant answers to RA-related questions.
• Link directly to relevant procedural documents.
• Assist in conflict resolution, key checkouts, and emergency protocols.
• Maintain an empathetic, supportive tone in interactions.

2. Data Sources and Curation

Dormibot pulls its knowledge from a curated set of publicly available Google Docs provided by the university’s housing department. These documents contain:

• RA policies and procedures.
• Contact lists for emergency and administrative services.
• Training materials and guidelines.
• Incident response protocols.

To ensure relevant and accurate responses, Dormibot utilizes Retrieval-Augmented Generation (RAG). The chatbot retrieves relevant document excerpts based on user queries and integrates them into its responses, improving precision and reducing hallucinations.

3. Features and Tools

• Retrieval-Augmented Generation (RAG): Fetches relevant text from RA policy documents to answer queries.
• Empathetic Response System: Dormibot maintains a warm, supportive, and understanding tone.
• Hostile Message Handling: If an RA is frustrated or upset, Dormibot de-escalates the situation with patient and constructive responses.
• UI Enhancements:
  • Favicon updated to a lantern icon to reflect the theme of guidance.
  • AI logo redesigned to match branding.
  • Custom fonts applied to improve readability and aesthetic appeal.
• Customizable System Prompts: The AI adapts its responses based on the context and nature of user interactions.

4. Code Modifications

Dormibot was built upon the myAI codebase but has undergone significant changes, including:

• Updated Default Response Messages:
  • New friendly and engaging introductory message.
  • Redesigned fallback message for when Dormibot cannot retrieve an answer.
• Overhauled UI.ts:
  • Updated colors, fonts, and branding.
  • Adjusted spacing and element positioning for better usability.
• Rewritten Prompts.ts:
  • New system messages tailored for RA-specific needs.
  • Specialized prompts for handling questions, conflicts, and support inquiries.
• Updated AI Identity:
  • Dormibot is now known as "Sean," reinforcing its approachable personality.
• Enhanced Search Logic:
  • Improved how document excerpts are retrieved and ranked for relevance.

5. Custom Prompting Strategy

Dormibot’s responses are guided by carefully designed system prompts, ensuring:

• Precision: AI provides direct and relevant information from RA policies.
• Empathy: Responses maintain warmth, humor, and supportiveness.
• Guidance-Oriented Approach: The AI directs users to specific documents via clickable links.
• Error Handling: If no relevant data is found, Dormibot acknowledges the limitation and suggests alternatives.

6. Access and Deployment

Dormibot is currently deployed at Dormibot Vercel. The full source code is available at GitHub Repository.

7. Future Improvements

• Implementing better search ranking for multi-document queries.
• Adding voice-to-text capabilities for mobile accessibility.
• Expanding knowledge sources beyond Google Docs.
• Enhancing response personalization based on RA experience levels.

---

This document serves as the technical overview of Dormibot, detailing its development, capabilities, and areas for future growth.