Chip Huyen

Profile

Chip Huyen is the person most developers end up reading when they stop asking “how do I train a model?” and start asking “how do I ship one?” She is the author of two of the most widely read practical AI books in the industry — Designing Machine Learning Systems (2022) and AI Engineering (2025) — and the second of those was the most-read book on O’Reilly’s platform in the year it launched. That is not a marketing statistic so much as a diagnosis of where the field is: thousands of engineers with solid software backgrounds needed a map of the foundation-model stack that wasn’t a vendor pitch, and she wrote it.

Her career reads like a deliberate tour of every layer where AI actually breaks. She was a core developer on NVIDIA’s NeMo, did ML research at Netflix, worked on data-centric tooling at Snorkel AI, and taught CS 329S: Machine Learning Systems Design at Stanford — the course that became Designing Machine Learning Systems. In 2022 she co-founded Claypot AI, a real-time feature engineering platform, which was acquired by Voltron Data in January 2024. There she served as VP of AI & Open Source Software, working on GPU-accelerated distributed data processing. As of 2026 she has moved on again and is building a new stealth startup.

What makes her worth reading, as opposed to merely worth following, is her refusal to let abstraction outrun evidence. The blog post “What I learned from looking at 900 most popular open source AI tools” is exactly what it says — she scraped the ecosystem and reported what was actually there, including the unflattering parts about how many projects are thin wrappers. “Building A Generative AI Platform” is a reference architecture that engineers have quietly copied into design docs at hundreds of companies. Her 8,000-word essay on agents arrived while the term was still mostly vibes and gave it a usable definition. She writes about evaluation — the genuinely hard, genuinely unglamorous problem — more than anyone at her level of visibility, and AI Engineering devotes serious space to it while most content on the topic skips straight to RAG diagrams.

For a developer learning AI today, the practical value is this: she draws the line between what you should build and what you should buy, and she keeps redrawing it as models get better. Her framing — start with prompting, escalate to RAG, escalate to finetuning only when you’ve proven you need it — has become close to default industry practice, largely because she articulated it clearly before it was obvious. Alongside authors like Aurélien Géron and Sebastian Raschka, she occupies the narrow and valuable space of people who write technical books that survive contact with production. Where they teach you the models, she teaches you the system around them.

Books

AI Engineering: Building Applications with Foundation Models

2024 ●

The definitive practical guide to building on top of foundation models — evaluation, prompt engineering, RAG, finetuning, inference optimization, and architecture — and the most-read book on O'Reilly's platform in 2025.

Designing Machine Learning Systems

An Iterative Process for Production-Ready Applications

2022 ●

An Amazon #1 bestseller in AI, translated into 10+ languages, covering the full lifecycle of production ML systems from data engineering to monitoring and drift.

📖

Machine Learning Interviews Book

A free, open-source guide to the ML interview process, covering both the technical questions and the career mechanics around them.

Key Articles & Papers

Agents 2025 — An 8,000-word treatment that gave 'agent' a working engineering definition — planning, tool use, failure modes — while the term was still mostly marketing. Common pitfalls when building generative AI applications 2025 — The mistakes she keeps seeing teams make: using AI where it isn't needed, skipping evaluation, and confusing demos with products. Building A Generative AI Platform 2024 — A layered reference architecture for GenAI systems — context construction, guardrails, routing, caching, observability — widely copied into real design docs. What I learned from looking at 900 most popular open source AI tools 2024 — An empirical survey of the open-source AI ecosystem, honest about how much of it is duplication and abandonware. Generation configurations: temperature, top-k, top-p, and test time compute 2024 — The clearest explanation available of the sampling knobs every LLM developer touches but few actually understand. Multimodality and Large Multimodal Models (LMMs) 2023 — A ground-up walkthrough of how CLIP, Flamingo, and multimodal architectures actually fuse vision and language. Open challenges in LLM research 2023 — Ten unsolved problems — hallucination, context length, multimodality, efficiency — that functioned as a research agenda for the field. RLHF: Reinforcement Learning from Human Feedback 2023 — A readable dissection of the three-phase pipeline behind ChatGPT-style models, written when most explanations were either hand-waving or papers. Building LLM applications for production 2023 — The essay that arguably named the discipline — prompt engineering's brittleness, control flow, and the gap between demo and production. Real-time machine learning: challenges and solutions 2022 — Online prediction versus batch, and the streaming infrastructure required to make continual learning real — the thinking behind Claypot AI.

Videos

Spotify Podcasts

YouTube