The development of GenAI applications based on company-owned data using Retrieval Augmented Generation (RAG) usually requires a multi-stage and complex process.

The first step is to select the appropriate search algorithm based on the data structure, which may also require data preparation. Choosing and efficiently using embedding models plays a crucial role in achieving useful embeddings; in some cases, these models even need adaptation. Once selected, the LLM model must be seamlessly integrated with the retrieval system, with factors like response speed and scalability being crucial.

After the basic architecture is in place, further aspects such as fine-tuning, prompt engineering and proper error handling need to be taken into account for productive use. Compliance and security requirements play an important role in deployment. It must be ensured that the application only provides authorized information in accordance with the user authorizations.