Automate LLM Testing with GPT-4 Judge Google Sheets Tracking

## 3. Judge LLM outputs
Our prompt judges the LLM input/output and decides if the LLM passed the test, based on how well the output fits to the reference answer. 

We also ask for a reason why the judge made its decision, which we can use to refine our eval later.

We're using OpenRouter here, which lets us easily tweak which LLM we want to use.

The output parser makes sure that the output is in JSON format, making the data easy to parse in the next step.