Proof-of-Concept
Sunbird Assistant

1. Introduction

This proposal outlines a simple Proof-of-Concept (PoC) for the Sunbird Assistant, an AI-powered tool that scrapes key information from the Fresno Pacific University (FPU) website and makes it accessible through a conversational interface powered by OpenAI GPT-4. This lightweight PoC focuses on validating the feasibility of extracting and presenting information from FPU's website via an intuitive chatbot.

2. Objectives

1. Demonstrate Feasibility: Validate the integration of a web scraper and GPT-4 to deliver website information conversationally.

2. Simplify Access: Provide students, faculty, and staff with quick, natural language access to publicly available university resources.

3. Lay the Groundwork: Establish a foundation for scaling the Sunbird Assistant for future advanced functionalities.

3. Scope of the PoC

3.1 Core Features

• Web Scraping:
  Scrape publicly accessible content from the FPU website (e.g., admissions, academic programs, contact information, and campus events).

• Conversational Access:
  Use GPT-4 to answer questions by referencing scraped content.

• Deployment:
  A simple web-based interface where users can interact with the assistant.

3.2 Exclusions

• Authentication or personalized responses.

• Dynamic or real-time website scraping during user interaction (scraping will be pre-scheduled).

• Integration with other university systems.

4. Technical Approach

4.1 Architecture

• Web Scraper:

  • Built using Python and BeautifulSoup or Scrapy for extracting structured content from the FPU website.

  • Periodic updates (e.g., daily) to ensure content freshness.

• Data Storage:

  • Store scraped content in a JSON file or simple database (e.g., SQLite).

• AI Engine:

  • OpenAI GPT-4 fine-tuned with instructions to refer to the scraped data.

• Frontend:

  • A lightweight web interface built using Flask or FastAPI for user interaction.

4.2 Flow

1. Scrape FPU website data.

2. Store data in a structured format.

3. Feed relevant data to GPT-4 based on user queries.

4. Generate and display conversational responses.

5. PoC Timeline

Phase

Tasks

Timeline

1. Planning

Define scraping scope, finalize GPT prompt design.

Week 1

2. Development

Build scraper, create basic chatbot interface.

Weeks 2-3

3. Testing

Test scraper accuracy and conversational output.

Week 4

4. Demonstration

Present working PoC and gather feedback.

Week 5

6. Budget Estimate for PoC

n/a

7. Success Criteria

1. Accurate Data Retrieval: Scraper successfully extracts relevant information from the FPU website.

2. Effective Responses: GPT-4 provides accurate and relevant answers based on the scraped data.

3. Positive User Feedback: Initial feedback from stakeholders indicates ease of use and utility.

8. Expected Outcomes

• A functional prototype demonstrating conversational access to FPU website information.

• Proof of technical feasibility for web scraping and GPT-4 integration.

• Insights into user expectations for future development phases.

9. Conclusion

This simple PoC for the Sunbird Assistant will validate the potential of combining web scraping and GPT-4 to improve access to university information. By focusing on publicly available content, this project lays the foundation for future enhancements while delivering immediate value through an intuitive conversational interface.

Would you like assistance creating the scraper or drafting GPT prompt instructions?