Is webllm + mistral 7b good for ICP frontend canister?


I want to implement AI in my frontend canister, and I decided to choose WebLLM + Mistral 7B because:

  1. it is lightweight
  2. private
  3. good performance for text-to-text

I am using it to convert text to JSON. For example:

Input:
“Create a 30 minutes event tomorrow 9 AM”

Output:

{
  "feedback": "please let me know the purpose (title) of meeting and who do u want to meet with",
  "attendees": [],
  "startTime": "2025-05-04T09:00",
  "endTime": "2025-05-04T09:30",
  "title": "30 minutes call"
}

Also, I may have more complex tasks like:

Input:
“Make a post about me to help me find a new job as ICP Rust/TypeScript+React developer”

Output:

{
  "feedback": "You you please add your work history like your GitHub? so people can learn more about you",
  "title": "ICP rust developer",
  "skills": ["rust", "icp", "typescript", "javascript"], // note AI must be smart to infer javascript despite it is not mentioned.
  "experiences": [],
  "salary": []
}