3. Generative process enhanced by information retrieval

Implementing a RAG approach requires the combination of a large number of software components and models, which need to be carefully selected according to their technical characteristics, the use cases envisaged and the constraints associated with them. Criteria to consider include :

• the nature, volume and confidentiality of information sources (documents and knowledge bases, or the Web);

• the adequacy of available models and the need to refine them on new data;

• the desired level of interactivity and availability (response time, scaling).

The deployment of a RAG solution is associated with decisive architectural choices, and a fine-tuning phase that can be complex, given the large number of strategies and parameters involved.