Create API function calling module for scanpy tl

[mengerj]

We will work on creating a structured query for an LLM, including the functions and parameters of the scanpy tools module to allow an LLM to run scanpy tools based on a user request.
@vd-dragan21
@bastienchassagnol

[mengerj]

Create generic method to generate a pydantic class for each function in a module: biocypher/scverse-x-biochatter#1

• Input: any python module
• Output: A list of pydantic classes of all functions (that don't start with "__") available in the module

[mengerj]

Example usage of the generic method with scanpy: biocypher/scverse-x-biochatter#2

Problems: expected methods that dont follow a function call with parameters like:adata.var["mt"] = adata.var_names.str.startswith("MT-")

[mengerj]

Next steps:
Integrate into biochatter framework

• include the generic function somewhere
• create an api_agent module for each desired module (eg. sc.pp, sc.tl,..) that hard codes the module as input to the generic function
• include the llm.build_tools() etc. into QueryBuilder.parametize_query method
• write unit tests
• Try (also as a test or just interactive?) different prompting extends

[mengerj]

optional scanpy depandancy, to ensure that user won't have to import scanpy anytime langchain is loaded.