judge Devpost submissions through AI's lens

                        
const message= [
{
    "role": "system",
    "content": `You are a hackathon judge. You will be given the hackathon requirements (what needs to be built)
    and prize criteria under which you need to declare a winner. You will then be given descriptions of two 
    projects along with their links.if anyone of the projects has no description return the other which has a 
    project description. incase none of them have any description return any one of them.give output strictly in 
    JSON format as shown below.

    ### SAMPLE OUTPUT
    {"Prize criteria": "Best Real-World App","Winner": "https://devpost.com/software/deverywhere","Reason": "justify the winner with chain of thoughts"}`
},
{
    "role": "user",
    "content": `
    The prize criteria under which you need to select the winner is: "Best Real-World App"
    ###

    Here is the overview of the first project:
    {
        "link": "https://devpost.com/software/wheely-wonka",
        "description": "Hi everyone, my name is Nico, and this is Spence GPT, my submission for the Google Chrome 
        Built-in AI Challenge. LLMs are known for their most boring use case: chatbots. But beyond that, 
        LLMs can do so much more. With prompt engineering, we can tweak each interaction exactly the way we want it, 
        and we can even achieve things like function calling, where the LLM decides if it is time to trigger a certain function. 
        Unfortunately, the prompt API is not yet ready to do classic function calling. But that got me thinking: 
        Can we achieve function calling with only a very structured way of prompt engineering? Turns out we can. 
        With Ben GPT, I wanted to create a car that I can actually talk to. At its core, I am using the Web Speech API and 
        also the Speech Synthesis API. With the Web Bluetooth API, I created a car that can be controlled with Ben GPT. 
        This project connects many browser APIs together and tests the limits of what is possible, inspiring creativity."
    }
    ###

    Here is the overview of the second project:
    {
        "link": "https://devpost.com/software/deverywhere",
        "description": "Hello everyone, my name is Joe, and for this hackathon, I created the Doagram Chrome extension. 
        It uses Chrome's built-in AI to create highly interactive diagrams of the concepts on the page you're looking at. 
        First, the extension breaks the web page into sections and passes each section to the Summary API, which creates a summary. 
        Then, I pass the summary to the Prompt API, which generates relationships between entities, and convert those into mermaid diagrams..."
    }`
},
{
    "role": "assistant",
    "content": `{
        "Prize criteria": "Best Real-World App","Winner": "https://devpost.com/software/deverywhere",
        "Reason": "BenzGPT utilizes the Speech-to-Text API to understand voice commands and uses the Prompt API to trigger functions for 
        controlling a virtual animated car. It also employs Text-to-Speech for feedback. While it is creative, it primarily serves as a recreational
        project without addressing a specific real-world problem. In contrast, Doagram creates flow diagrams from web page content, making it useful 
        for individuals who prefer learning visually. Given its practical applications, Doagram is the winner for the 'Best Real-World App' category."}`
},
{
    "role": "user",
    "content": `
    The prize criteria under which you need to select the winner is:
    ###
    ${prize}
    ###

    Here is the overview of the first project:
    {
        "link": "${link1}",
        "description": "${des1 || "no description"}"
    }
    ###

    Here is the overview of the second project:
    {
        "link": "${link2}",
        "description": "${des2 || "no description"}"
    }`
}];