Develop Starsim-specific AI copilot

[cliffckerr]

To help users learn how to use Starsim, or to migrate between major versions, would be nice to have a multi-agent RAG LLM copilot.

First use case is to migrate from Starsim v1.0 to v2.0. Conceptually I'm imagining it would work as follows:

• Input Starsim v1.0 source code and examples as context (or as a RAG)
• Likewise with Starsim v2.0 source code
• Select a file or files to translate/migrate from v1.0 to v2.0
• Perform the migration
• Optionally: have a set of expected outputs from v1.0 code, and have a different agent check that v2.0 outputs match (or are close to matching) these.

Imagining an API like:

 import llm_migrate as llmm

llm = llmm.create(model='some-big-LLM-like-gpt-4-32k') # Connect to e.g. OpenAI API

llm.set_context_from('/path/to/old/code') # Set context/RAG for the "old" (v1.0) version
llm.set_context_to('/path/to/new/code') # Set context/RAG for the "new" (v2.0) version
llm.set_migrate('/path/to/files/to/migrate') # Select the folder/files to migrate
llm.set_target('/path/to/expected/outputs') # Optional; select the folder with output for validation
llm.migrate() # Perform the migration
llm.validate() # Optional; validate the output of the migrated files against the target outputs

[cliffckerr]

Ideas/resources:

• https://github.com/krosenfeld/bumper_paper/blob/main/bumper/bumper/base.py
• https://blog.withmantle.com/code-conversion-using-ai/ (this one looks good!)
• https://medium.com/@kapoorchinmay231/large-language-models-llms-for-code-conversion-new-age-of-ai-72ebd2c8918d
• https://medium.com/@gil.fernandes/a-cobol-to-python-converter-with-autogen-agents-7f4db7455ee1 / https://github.com/onepointconsulting/cobol-converter