Using microservices to secure agentic LLMs

[chbrooks]

I'm hoping to get a student started on this:

• the idea: take the Google Earth Engine tool that Simon created, identify the places where the LLM is directly executing code, and replace those with calls to microservices, where the security surface can be articulated more clearly.
• Proof of concept - can we then get an LLM to generate microservice calls that replicate the original functionality, but with improved security?