It’s possible to do this either by fine-tuning an existing model or using an existing chat model prompts enriched by a vector search for relevant code. See my comment elsewhere.