LLM Augmentation

Techniques that enhance LLM capability at inference time without modifying model weights. The model stays frozen; augmentation comes from what enters the context, what tools are available, how reasoning is orchestrated, and how output is controlled

Modifying a foundation model's weights to encode domain-specific knowledge, behavior, or capabilities

Designing and constructing the structured knowledge artifacts that runtime augmentation mechanisms consume. The substrate is the foundation layer — taxonomies, graphs, schemas, and rules that make retrieval efficient (category-first, then data), prompting precise, and reasoning constrained