[2/2] Hands-On RAG for Production (AI column)

Finishing. survey it's cool books RAG, which is in the process of writing.

5) "Evaluating your RAG Application" This chapter is a must read for those who plan to finish RAG production. The authors mention metrics such as hallucinations, response quality, latency and cost. And not only mention, but also talk about ways to measure the quality of retrieval and generation. In fact, here is a story about standard search metrics. (Precision, recall, f1, metrics based on order of elements)and the accuracy of generation (utilization of context, accuracy of answers, consistency of answers, absence of hallucinations, accuracy of quotations)and prejudices (race, sex, etc.). As well as approaches to e2e are evaluated using frameworks: Open-RAG-EVAL, RAGAs, DeepEval. And finally, how to take into account the feedback of people (Conventionally, the fingers up and down that you saw in chats ChatGPT, Perplexity.)

6) "From RAG to AI Agents" Retrieval is no longer the end product and becomes part of a longer workflow. That is, RAG is not just “find and answer”, but “find, check, plan, call the tool, return the result”. This is very true for the current time, where agents without quality retrieval quickly turn into expensive improvisation.

7 and 8) Multimodal RAG and Knowledge Enhanced RAG This book is even more useful for real companies. This is important for anyone who works with PDF, tables, diagrams, images, semi-structured documents and complex knowledge domains, where embedding similarity is no longer enough.

P.S. If you compare this book to the others I’ve talked about, it’s like this.

• "AI Engineering" This is a broad map of the entire discipline and the answer to the question, what is modern AI development and what layers it consists of. Against this background, Hands-On RAG for Production no longer looks like an overview of the entire system, but like a very detailed analysis of its part - production RAG.
• "Prompt Engineering for LLMs" Learns to understand the LLM architecture, build a prompt strategy, properly assemble context elements, and use techniques like few-shot, chain-of-thought, and RAG. Therefore, Prompt Engineering for LLMs is a book about the interface between human, context and model, and Hands-On RAG for Production is about the retrieval/platform layer that this context makes reliable in the product.

#AI #Engineering #Software #DistributedSystems #SystemDesign #Database #Search #Agents