Skip to content
#AI

AI engineering. Building applications using basic models (AI Engineering) (AI column)

#AI #Architecture #Software #Engineering #ML #Data #SystemDesign #DistributedSystems

The bookAI Engineering- already transferred. release Peter's publishing house. I read this book in English at the beginning of the year and it was great. (We, along with Flo’s Zhenya Sergeev, almost completely dismantled it in three series of the podcast: 1, 2 and 3). Actually, the book is great for engineers who are building a GenAI product right now, for example, chatbot, copilot, RAG, etc.

She wrote a book. Chip Huyenwhich She taught Machine Learning Systems Design at Stanford. (CS329S)It can be explained systematically and easily.

  • I wrote a book.Designing Machine Learning Systems" (O’Reilly)A must-read for ML/Platform engineers Worked at the junction of AI + infrastructure (NVIDIA, Snorkel AI)Now - Voltron Data (GPU Acceleration of Analytics)

There are several reasons why AI Engineering became a bestseller.

  1. I'm in pain. 2024–2025Teams massively integrate basic models, and you need an engineering framework, not another set of prompts
  2. The big focus was on evaluation: not “vibe check”, but measurable metrics, regression sets, LLM‐as‐a‐judge, degradation testing.
  3. Closes the product reality: cost and latency of inferencing, monitoring, data quality / feedback loops, security and risk management
  4. It maps patterns. (prompting, RAG, fine‑tuning, tools/function calling, agents) The main question is when to choose.
  5. The book has public materials/resources (repo on 12.5k stars)The ecosystem of practices grew rapidly around it.

Here's what this book might be useful to engineers who are already sawing GenAI

  • Turns a demo on a framework into a project: requirements → architecture → eval harness → release → observability
  • Helps to speak the same language with Product/Platform: what SLO/SLI LLM features, what we log, where we do gate quality, as we consider cost-per-request Good for Tehlids: a lot about compromises (Quality vs Price, Context vs Retrieval, Determinism vs Creativity) and where the system most often “breaks”

A mini-checklist from a book that can be used tomorrow:

  • Golden cases + negative cases (Regular regression evals) Prompt/model/retriever versioning + context logging Metric "cost of response" + budget for the feature
  • Eval gates in CI before depo (as a unit/integration for LLM only)

The book is definitely worth reading. By the way, while the book is still in pre-order for it is a discount in 35% when using the promotional code "Pre-order"

#Architecture #Software #AI #Engineering #ML #Data #SystemDesign #DistributedSystems