BisonBot: Your Howard Information Guide
Overview
BisonBot is an AI-powered chatbot designed to enhance the user experience on Howard University's website. By addressing the challenges of navigation and accessibility, BisonBot provides a frictionless interface for students, faculty, and visitors to access information efficiently.
Currently, the website looks like this:
With the current setup, the bot is only accessible on a few web pages. But, with BisonBot we aim to make the feature available in a separate browser so users can ask Howard related questions and get responses.
Features
- Authentication: Users can sign in using GitHub and/or Google.
- Prompt Handling: Intuitive processing of user queries with response generation tailored to Howard's spirit.
- Accessibility: Toggle between light and dark themes, and read-aloud options.
Requirements
- Login and logout capabilities
- Responsive chat with history viewing and new tab initiation
- Accessibility features to support diverse user needs
Technologies Used
- GitHub OAuth: Ensures secure user authentication.
- Python and OpenAI API: For backend development and chat response handling.
- Chainlit: Enhance the front end and support real-time interactions.
Additional Features
- Third-Party Logins: Integration with platforms like Google and GitHub.
- User Identification: Custom responses based on user roles (students, parents, faculty)
- BisonBot Icon: Quick access chat window on all web pages.
Database Design
- Chat History Storage: Managed via Literal AI API.
- Vector Database: Chroma DB stores vector embeddings for efficient retrieval.
- User Information Storage: Managed by ChainLit’s Database.
User Interface
- Integration with GitHub/Google for login.
- Real-time chat interface with options for attachments and copying text.
- Toggle for light/dark mode.
Testing Strategy
- Predefined Benchmarks: Evaluate responses for accuracy, relevance, and tone.
- Prompt Variability Testing: Ensure diverse and contextually appropriate responses.
- Content Generation and Retrieval: Test for factual accuracy and up-to-date information.
Quick Start Guide
- Clone the repository and navigate to the project directory.
- Install dependencies:
pip install -r requirements.txt
.
- Access BisonBot through localhost on your browser.