Meet Sebastian Raschka

Sebastian Raschka and his portfolio website

Sebastian Raschka is an LLM research engineer, author, educator, and former statistics professor at the University of Wisconsin–Madison. His work focuses on large language models, reasoning models, deep learning, practical machine-learning systems, PyTorch, and “from scratch” implementations that explain modern AI systems through code.

His homepage positions him as an AI practitioner who bridges academia and industry, with prior work as a senior engineer at Lightning AI and academic research/teaching experience at UW–Madison. The site is both a personal portfolio and a large technical resource library for people learning machine learning and LLM engineering.

Main areas of work

Raschka’s current focus is strongly centered on LLMs:

• LLM architecture, pretraining, finetuning, inference, and evaluation
• Reasoning models, inference-time scaling, reinforcement learning, and distillation
• PyTorch-based implementation and performance-oriented AI engineering
• Technical education through books, courses, tutorials, blog posts, and open-source repositories

His About page emphasizes clear explanations, working code, and end-to-end examples as the core style of his work.

Books and learning resources

His Books page highlights several major publications:

• Build a Large Language Model (From Scratch) — a hands-on book and companion hub for building a GPT-style LLM in Python and PyTorch, covering embeddings, attention, transformer blocks, pretraining, and finetuning. The companion repository is prominently featured and has about 97k GitHub stars.
• Build a Reasoning Model (From Scratch) — an in-progress/early-access Manning book focused on adding reasoning capabilities to pretrained LLMs, including evaluation, inference-time scaling, RL-style training, and distillation.
• Machine Learning Q and AI — a question-and-answer style book on machine learning, deep learning, and AI topics.
• Machine Learning with PyTorch and Scikit-Learn — a broad practical ML/deep-learning book using scikit-learn and PyTorch.

The site also includes course materials for LLMs, PyTorch, deep learning, and machine learning, including a free LLMs-from-scratch YouTube course, PyTorch tutorials, and materials from his UW–Madison courses.

Blog, notes, and LLM research tracking

The blog and notes archive is one of the central parts of the portfolio. Recent writing is heavily focused on LLM research and engineering, with topics such as:

• LLM architecture comparisons
• attention variants, KV cache, sparse attention, and MoE models
• open-weight model releases such as Qwen, DeepSeek, Gemma, Llama, GPT-OSS, and others
• reasoning LLMs, inference-time scaling, RLVR, and evaluation
• from-scratch implementations of tokenizers, self-attention, LoRA/DoRA, Qwen, and LLM evaluation methods

The blog functions as a technical research notebook as much as a conventional personal blog.

Notable portfolio projects

A standout resource is the LLM Architecture Gallery, a curated reference of modern language-model architectures. It includes model cards, diagrams, fact sheets, implementation links, and a comparison/diff tool for dozens of LLMs. It reflects his broader emphasis on making fast-moving AI architecture research easier to understand.

His software page lists selected open-source projects, including:

• MLxtend — a Python library of machine-learning utilities
• CORAL and CORN — ordinal-regression methods for deep neural networks
• BioPandas — molecular-structure data handling with pandas DataFrames
• ScreenLamp, SiteInterlock, and related computational-biology tools
• utilities such as watermark for reproducible research metadata

Research background

The Publications & Research page shows a broad research history spanning deep learning, ordinal regression, model evaluation, face-privacy methods, few-shot learning, computational biology, protein–ligand modeling, and machine-learning software. Earlier work includes JOSS papers for MLxtend and BioPandas, papers on semi-adversarial networks for face privacy, ordinal regression for neural networks, and ML for GPCR ligand discovery.

Talks and public presence

His Talks and Events page lists many invited talks, tutorials, workshops, and keynotes at venues such as PyTorch Conference, SciPy, PyCon, NeurIPS workshops, CVPR tutorials, and ACM Tech Talks. Recent talks focus on LLM development, LLM architectures, reasoning models, PyTorch, finetuning, and “from scratch” learning paths.

You can find him on GitHub, LinkedIn, X, YouTube, Google Scholar, and his Substack-style publication Ahead of AI.

Overall impression

Sebastian Raschka’s portfolio is less a simple résumé site and more a comprehensive technical learning hub. It presents him as a hands-on LLM researcher and educator whose niche is translating frontier AI concepts into readable explanations, PyTorch code, diagrams, books, courses, and open-source repositories. His recent work is especially concentrated on understanding and implementing modern LLMs and reasoning models from first principles.