The biggest challenge with ML models isn't the algorithm but the organization of contextual knowledge. In my experience, hierarchical structuring of documentation significantly improves results, especially when working with LLMs.