At the mometn all prompts are stored in the prompts.py file. The prompts are mostly optimized for the Gemini models (2.5-pro and 2.5-flash). However, different models respond better to different prompting styles: https://promptbuilder.cc/blog/claude-vs-chatgpt-vs-gemini-best-prompt-engineering-practices-2025

Goal:

Have an abstract factory which will provide all prompt strings for everything in the codebase that is:
Tool names, Tool descriptions, System prompts, User prompts, Extractor prompts

Implementation:

• AbstractFactory pattern with a default factory which is set to the current prompts.
• The factory has to select the LLM alongside with all of its strings
• The factory itself has to be a singleton which is used by all the different components around the codebase.
• Unit tests that the default factory is selected when the instanitated llm has no prompt implementation yet.
• Have a nice prompt structure - can be outside of the *.py, can be something like a properties file or something else relevant for the task,