Semantically sanitize user information when prompting Gemini

[hoanghm]

Like SQL Injection, users may use malicious information to generate unexpected missions (e.g., ridiculous, irrelevant tasks)

• When user information is filled out, they need to be sanitized semantically to avoid "prompt injection"

• [ ] 1. Experiment with prompt templates to sanitize input, then create a method for GeminiClient to sanitize an input given:

  • Input field name (e.g. interest)
  • Expected answer (e.g., an interest from this user)

• [ ] 2. Set up an API endpoint to call and sanitize inputs semantically for the front end.