[Feat] : Security system for LLM

[iam-vipin]

What do we need?

Right now, anyone can type anything into the prompt and receive a response.

This should not happen—AI should not reply with anything inappropriate or unprofessional.

Acceptance Criteria

• Add a prompt sent with all AI-related API calls.
• Create a new prompt to ensure that AI does not respond to inappropriate inputs but provides a common response in such cases.
• Handle all other cases where the LLM may generate incorrect or undesired responses.

Relevant ScreenShots

No response

Further Comments

No response

[not-meet]

Hey @VipinDevelops can you assign this issue to me I would like to contribute to this by enhancing the security measures by tweaking the prompts

[iam-vipin]

Thanks for your interest! I prefer to give it to someone new. Those who already have an assigned issue should focus on getting those merged first

[sushen123]

hey @VipinDevelops can i assigns this to me?

[thekishandev]

Hi @VipinDevelops

I'd like to work on this security system for the LLM integration. here's my approach:

1. I'll create a "security wrapper" around all LLM API calls that includes:

   • system prompt that instructs the AI to refuse inappropriate requests
   • Input validation to catch obvious problematic patterns before they reach the API
   • Response filtering to ensure we never display unsafe content

2. For implementation, I'm thinking of:

   • Creating a middleware function that processes all user inputs
   • Adding configurable security levels (strict/moderate/minimal)
   • Building a small library of standardized rejection messages that feel natural

I've worked with LLMs before and understand the challenges of prompt engineering for safety. Happy to adjust my approach based on your feedback!

Can I get assigned to this issue? I don't have any other open issues and am excited to contribute to improving the app's security.

Thanks!

[moyrsd]

Can I work on this?

[iam-vipin]

@moyrsd Sure, Share the approach here onc, and you can confirm with @Kishan-Patel-dev if he is still working on it.

[thekishandev]

@VipinDevelops waiting for your response I have shared my approach.

[iam-vipin]

Oh okay Yeah you can go ahea with implementation.

[thekishandev]

@amitb0ra I am working on this issue you can check other issues in this repo

[thekishandev]

@VipinDevelops Can please review the PR implementing #61 Security systems for LLM