AI hint-Prompt needs an improvement

[decodingafterlife]

📌 Description

The current prompt for hints is:

  try {
      const response = await fetch('https://api.groq.com/openai/v1/chat/completions', {
        method: 'POST',
        headers: {
          'Content-Type': 'application/json',
          'Authorization': `Bearer ${apiKey}`,
        },
        body: JSON.stringify({
          model,
          messages: [
            {
              role: 'system',
              content: `You are an AI assistant for LeetCode. Provide a ${level} hint for the following problem. Do not solve the entire problem. Just provide a hint.`,
            },
            {
              role: 'user',
              content: context,
            },
          ],
        }),

This is very basic and has lot of flaws:

• No consistency in the generation of hint (The same type of hint can result into two different responses from the same model).
• No consistent hierarchy or improvement of hint, sometimes the ordinary hint can be better than advanced even if it costs less.
• Bad prompt.

---

✅ Tasks

1. Overall improvement in hint generation:

• Overall improvement in prompt structure.
• Create three distinct prompts tailored to the three hint tiers: ordinary, advanced, and expert. Each should build upon the previous level in depth and technicality.

2. Improvement in the consistency of hints:

• Use temperature parameter to keep the hints more balanced and consistent.

3. LangChain:

• Use of LangChain to store the 3 different prompts and send them to the model based on the requested hint from the user.
• Add a word limit so that the hint is not too long or too short.
• Currently when we use the deepseek model we also see the part of the responce. Langchain can be used to remove this so that only the hint is displayed.

4. Testing:

• Testing all that is mentioned above.
• Run tests comparing responses from each hint level for multiple problems to check for hierarchy consistency.

---

📂 Relevant Files

• `popup.js'
• 'prompts.js': Maybe create a new file for storing the prompts and the LangChain workflow.

🔴 Currently a level2 contribution but can be upgraded to level3 on proper resolution.

[riyaa404]

Hi @decodingafterlife I would like to contribute to this issue.
I’ve started exploring the improvements, Kindly assign this to me.

[decodingafterlife]

@riyaa404 can you please elaborate on how you are planning to work on this.

[riyaa404]

Here's my plan :
I’ll create structured and different prompts for Ordinary, Advanced, and Expert hint levels.
Instead of direct fetch, I’ll use LangChain JS to manage prompt templates, handle model responses, and apply formatting.
I’ll add word limit caps per hint level , For ex : 50/100 words etc.

[decodingafterlife]

sounds good, do test your prompts and add share the results in the comments of the pill request through ss or text. also test the word limit, which one gives the best result. Also set the temperature parameter low.

[decodingafterlife]

@riyaa404 update?

[riyaa404]

Created prompt.js and updated popup.js files.
Done with UI setup for ordinary, advanced, and expert levels with hint system and HTML popup and updated dependencies.
Pending: fix for hint visibility and final testing.

[anshiagrawal22]

@riyaa404 any update?

[riyaa404]

Hi @anshiagrawal22
I have recreated the pull request with all the improvements suggested. The merge conflicts have also been resolved

The PR was already recreated three days ago but at that time I mistakenly left out the description, which also resulted in the issue not being linked. I have now updated the PR with a complete description, properly linked it to issue and ensured everything is in place for review.