Jay Alammar Portfolio
Director and Engineering Fellow at Cohere
Creator of "The Illustrated Transformer" and "The Illustrated BERT" — visual explainers of core LLM architecture that became widely used references for understanding how transformer models actually work; co-author, with Maarten Grootendorst, of the O'Reilly book Hands-On Large Language Models and its free DeepLearning.AI short course, "How Transformer LLMs Work." The blog has been frozen since March 2025, with new writing moved to Substack.
Location
Coordinates unknown. Vibes: international.
Companies
Skills
Notable
- Co-author of Hands-On Large Language Models (O'Reilly); the companion code repo has 29.3k GitHub stars
- Co-instructor of How Transformer LLMs Work, a DeepLearning.AI short course
- Creator of Ecco, an open-source library for visualizing Transformer language models (2.1k GitHub stars)
- The Illustrated Transformer says it has been featured in courses at Stanford, Harvard, MIT, Princeton and CMU
What makes it work
A breakdown of the choices that make this portfolio stand out.
One consistent format applied to hard technical concepts
Custom diagrams paired with plain-language explanation is the same format across every major post, applied to genuinely difficult internals (attention, embeddings) rather than surface-level topics.
A body of work built entirely on a static site
The blog runs on GitHub Pages with no CMS overhead — proof that technical depth and visual polish matter more than platform sophistication for building a reference resource.
Carrying the format into a book and a free course
The Illustrated Transformer post now points readers to Chapter 3 of Hands-On Large Language Models (O'Reilly, almost 300 custom figures) and to a free DeepLearning.AI short course that animates the same material. The line from free explainer to book is visible, and the underlying craft never changes.
What AI Engineers can take from this
Specific, actionable tips to apply to your own portfolio — no generic advice.
Develop one distinctive explanation format (a diagram style, a metaphor, a structure) and apply it consistently to your hardest technical topics.
Don't wait for platform sophistication — a static site with genuinely good content outperforms a fancier site with generic content.
When free content builds a real reputation, carry the same format into a longer, structured product, such as a book or course, instead of starting a different approach from scratch.
Projects
The blog has been frozen since March 2025, and it barely matters: Jay Alammar's illustrated posts still carry update banners pointing to the book chapter and course that grew out of them. An archive that forwards its readers to newer versions is a rare kind of maintenance.
The Illustrated Transformer
A visual, step-by-step breakdown of the Transformer from "Attention is All You Need": encoder and decoder stacks, self-attention, multi-head attention, positional encoding and the training loop, each introduced one concept at a time with custom diagrams. It carries its own academic citation format (Alammar, J (2018)), and translations are linked.
Explaining attention clearly enough that universities assign it is a rare skill. For AI engineering roles, it shows a first-principles understanding of the architecture being deployed.
Introduces the model as a black box, then opens one component at a time
Update banners route readers to the newer book chapter and course
Published under CC BY-NC-SA with a suggested citation
Hands-On Large Language Models (book and code)
"The Illustrated LLM Book", co-written with Maarten Grootendorst for O'Reilly, with almost 300 custom figures. It covers tokens and embeddings, how transformer LLMs work, semantic search, RAG and fine-tuning. Every chapter has a companion notebook built and tested on Google Colab's free T4 GPU.
It pairs the illustrated explanations with runnable code for each chapter, bridging understanding and implementation, which is what LLM engineering teams need.
Chapter notebooks run on free Colab GPUs
Almost 300 custom figures in the same visual style as the blog
Companion free course on DeepLearning.AI
A library to explain, analyze and visualize NLP language models. Ecco creates interactive visualizations directly in Jupyter notebooks that explain the behavior of Transformer-based models like GPT-2, BERT, RoBERTa, T5 and T0.
It turns the blog's visual intuition into tooling other researchers can run on their own models, making interpretability something you can do in a notebook.
Interactive in-notebook visualizations of model internals
Supports GPT-2, BERT, RoBERTa, T5 and T0
Documented on Read the Docs
More AI Engineer portfolios
Ready to build your portfolio?
Follow the AI Engineer roadmap — skills, projects, and timeline to get hired.