I’ve been having fun generating code via LLMs but getting frustrated with the common pitfalls (hallucinated functions, parameters, mixed up language constructs etc.). I use PowerShell a lot in my daily work so I built something to help generate a lot of boilerplate code I need. I’ve used the PS abstract syntax tree to parse the LLM output and validate that common issues aren’t present, if they are it gets auto-corrected and it also uses the LLM to ensure the logic is still sound after correcting syntax issues.
I probably spent more time writing the terrible in-terminal syntax highlighter than working on the AI side of things
