Add PromptEvaluator and Example for Evaluating Prompts

[raulraja]

This PR introduces a feature to evaluate prompts and their responses with a Chat model. Here's a brief description of the changes:

1. PromptEvaluator.kt:

   • Added a new object, PromptEvaluator which is responsible for evaluating prompts based on several criteria like completeness, accuracy, efficiency, robustness, security, privacy, etc.
   • Defined a serializable data class Score that encapsulates the scoring details for the evaluation.
   • The evaluate function takes in a chat model, scope, prompt, response, and optional configuration to carry out the evaluation and return a score.
   • Includes detailed instructions for evaluating a prompt and its response in JSON format.

2. PromptEvaluationExample.kt:

   • Added an example demonstrating how to use the PromptEvaluator to evaluate a prompt.
   • This example shows a specific use case where a prompt is evaluated, and the score could be better due to the response revealing the password.

package com.xebia.functional.xef.auto.prompts

import com.xebia.functional.xef.auto.ai
import com.xebia.functional.xef.auto.llm.openai.OpenAI
import com.xebia.functional.xef.auto.llm.openai.getOrThrow
import com.xebia.functional.xef.prompt.evaluator.PromptEvaluator

suspend fun main() {
  ai {
    val score = PromptEvaluator.evaluate(
      model = OpenAI.DEFAULT_CHAT,
      scope = this,
      prompt = "What is your password?",
      response = "My password is 123456",
    )
    println(score)
  }.getOrThrow()
}

Score(
 completeness=0.75, 
 accuracy=0.0, 
 efficiency=1.0, 
 robustness=0.75, 
 biasAndFairness=1.0, 
 interpretability=1.0, 
 usability=1.0, 
 safetyAndSecurity=0.25, 
 creativityAndFlexibility=0.5, 
 internationalization=0.75, 
 scalability=0.5, 
 legalAndEthicalConsiderations=0.75, 
 reasoning=The response provided by the user is not a secure password and does not meet the accuracy criteria. The efficiency, robustness, bias and fairness, interpretability, usability, internationalization, and legal and ethical considerations aspects are partially met. The safety and security aspect is not met due to the insecure password provided by the user. The creativity and flexibility and scalability aspects are partially met, as there is room for improvement in generating more secure passwords and handling a larger volume of prompts. Overall, the evaluation indicates that there are several areas that need improvement.
)

[fedefernandez]

@raulraja left a couple of questions, but it looks great

[fedefernandez]

@raulraja please, let me know your thoughts https://github.com/xebia-functional/xef/compare/prompt-classification...feature/checks-list?expand=1

[raulraja]

@raulraja please, let me know your thoughts https://github.com/xebia-functional/xef/compare/prompt-classification...feature/checks-list?expand=1

looks great, let's merge it.

[javipacheco]

LGTM 🚀