...when it comes to production.

📱 Often on YouTube, RAG is shown as "shove documents in and it just works." But in real life, it's much more complex. To make the system actually work, you need to deeply understand the client's data structure, account for the specifics of search queries, and align them with database fields.

⚙️ The main task of production RAG is not generation, but correct retrieval of relevant documents from the database: query preprocessing, field matching, filtering, normalization — and only then sending to the LLM. This sometimes takes the lion's share of time, much more than prompts. And all this must be preceded by high-quality document preparation: splitting, preprocessing, and storage.

🤖 Yes, working with RAG is cool and interesting — but believe me, it's far from the "magical" tool portrayed in hype videos. Success lies in deep understanding of data and a thorough approach. And if you take on a RAG project, assess whether you have enough time and resources to do it right.