An off-the-shelf LLM does not know your products, policies or documents. Two common ways to fix that are retrieval-augmented generation, known as RAG, and fine-tuning. They solve different problems.
What RAG does
With RAG, the system searches your documents for relevant passages and gives them to the model along with the question. The model answers using that material, and can point to its sources. It is well suited to knowledge that changes, such as policies, price lists and product manuals.
What fine-tuning does
Fine-tuning further trains a model on examples so that it adopts a style, format or specialised behaviour. It changes how the model responds rather than giving it a searchable library of facts.
How to choose
- Need answers from your own documents? Start with RAG.
- Information changes often? RAG is easier to keep current.
- Need a consistent tone, format or specialised task behaviour? Consider fine-tuning.
- Need both? They can be combined.
Start with the simplest option
Good prompting is the cheapest step, RAG is the next, and fine-tuning is usually last. Many business needs are met before you reach fine-tuning.
Quality depends on your data
Either method performs badly on outdated or messy documents. Cleaning and organising your content is often the most valuable part of the project.
We build knowledge assistants using this approach. See our AI & Automation page or contact us.