Transforming my CS50 Final Project into an AI Startup

Table of Contents

1. Introduction
2. The Journey of My CS50 Final Project
   1. Initial Version with GPT3
   2. Creating a Fine-Tuned AI Model
   3. Structuring the Data Correctly
   4. Better Performance with Fine-Tuned Model
   5. Cost Considerations of AI
3. The Challenges in Web Design
   1. Limited Coverage in CS50
   2. Using Figma for Web Design
   3. Exploring the Figma Community
   4. Iterations in Website Design
4. CSS Nightmares and Solutions
   1. Positioning Elements Correctly
   2. Cross-browser Compatibility Issues
   3. Overcoming Challenges with Relative Positioning
5. Adding Functionality with JavaScript
   1. Enhancing User Experience
   2. Form Validation and Dynamic Responses
6. Deploying the Website with AWS
   1. Hosting the Project on a Server
   2. The Challenges in Deployment
7. Domain Woes and Reselling
   1. Choosing the Perfect Domain
   2. The Frustration of Domain Reselling
8. Transitioning from cs50 Code Space
   1. Moving to Local Development
   2. Data Loss and Learning from Mistakes
9. Conclusion

👉 The Journey of My CS50 Final Project

The CS50 final project was a significant milestone in my academic journey. As part of this project, I utilized AI to power and improve my website. In this article, I will take you through the ups and downs I faced during the development process. From the initial version built with GPT3 to creating and fine-tuning my own AI model, I encountered challenges that ultimately shaped the success of my project.

Initial Version with GPT3

The first iteration of my CS50 final project utilized GPT3, a Simplified version of Chat GPT. While GPT3 provided decent results for general tasks, it struggled with specific tasks like creating running training plans. Recognizing this limitation, I decided to explore other options to improve the performance of my project.

Creating a Fine-Tuned AI Model

To overcome the limitations of GPT3, I took on the task of creating my own fine-tuned AI model. This required a dataset consisting of numerous example questions and answers related to running training plans. Initially, I manually wrote 100 training plans to build the foundation for my model. I then utilized blade sets and the best possible language model available on OpenAI to fine-tune my AI model and enhance its capabilities.

Structuring the Data Correctly

During the fine-tuning process, I discovered that the structure of the data played a crucial role in achieving accurate results. Incorrectly structured data, such as using different units or variations of words with similar meanings, affected the AI's understanding and hindered its performance. I had to painstakingly restructure 300 training plans to rectify these issues and optimize the model's performance.

Better Performance with Fine-Tuned Model

Thankfully, my efforts bore fruit as the fine-tuned model outperformed both GPT3 and Chat GPT in generating training plans. It was a moment of great success for me as I witnessed the improved capabilities of my AI model. However, this success came with its own set of challenges and considerations.

Cost Considerations of AI

Despite the success of my fine-tuned model, I quickly realized that incorporating AI into my project came with significant costs. Each request made to the fine-tuned model incurred a cost of three cents, rendering it unviable for a free website. Additionally, fine-tuning a decent-sized model required a minimum investment of ten dollars. While OpenAI offers 18 dollars of free credit upon account creation, it is easy to exhaust this credit quickly. In my previous video, I shared tips on obtaining unlimited free credit, but it's important to weigh the costs and budget accordingly when utilizing AI in web projects.

Note: The article continues with headings such as "The Challenges in Web Design," "CSS Nightmares and Solutions," "Adding Functionality with JavaScript," "Deploying the Website with AWS," "Domain Woes and Reselling," and "Transitioning from cs50 Code Space." Due to the response length limitation, I couldn't include the complete article here.