About
This graduate course treats fluency with generative AI as a research skill, not a novelty. Over fourteen weeks, engineering doctoral students build that fluency directly into their own research workflows: reading and synthesizing literature, writing and reasoning about code, and organizing what they know. This is not a computer science course and the goal is not to build AI systems, but rather to use them well, with enough understanding of how they work to use them critically.
The sessions move from foundations to practice. The first few establish how modern language models actually work—covering transformers and attention, reasoning, tokenization, and retrieval—and survey a landscape that shifts month to month. From there the course turns to doing: AI-assisted literature review, agentic coding with tools like Claude Code, and the conventions (MCP, skills, project files) that make agents genuinely useful for research. One idea connects all of it: context engineering, the deliberate design of what an AI system sees and remembers. The same discipline that shapes a good prompt also shapes a well-structured repository, a research notebook, and a personal knowledge base.
A second thread runs throughout: a habit of critical rigor. The course opens with the illusions of understanding that AI can foster in science, and closes by asking how these tools might sharpen rather than dull a researcher's thinking. Students are treated as peers, capable of weighing real limitations and of telling what AI demonstrates computationally from what it validates experimentally.
All materials are shared openly under CC BY 4.0 and evolved as the field did over the spring of 2026. They were not revised and fit their moment in time, but they are extensive. The schedule below links the lecture notes and slides for each session; the assignments tie every technique back to students' own research.
Details
- Instructor
- Prof. Lorena A. Barba
- Prerequisites
- PhD student status in engineering (or permission of instructor) and proficiency in Python; no prior AI or machine-learning coursework required.
- Contact
- labarba@gwu.edu
| Class | Date | Topic | Notes | Slides |
|---|---|---|---|---|
| 1 | Jan 13 |
Course introduction; AI-for-science overview; the illusion of knowledge
Read:Messeri, L. and Crockett, M.J., 2024 (PDF) Watch:The Thinking Game; Generative AI in a Nutshell (explainer in 18 min) |
Notes | Slides |
| 2 | Jan 20 |
Foundations of generative AI: neural networks, transformers, attention; reasoning, chain-of-thought; context, tool use, RAG
To do:"How Transformer LLMs Work" on DeepLearning.AI; Assignment 1 (by next class) Watch:3Blue1Brown video course |
Notes | Slides |
| 3 | Jan 27 |
Current LLM landscape; open models; reasoning models deep dive; long context; mixture-of-experts
Read:Generative AI's Act o1: The Reasoning Era Begins. Sequoia Capital, October 9, 2024. |
Notes | Slides |
| 4 | Feb 3 |
AI-enhanced literature review; semantic discovery tools; deep research agents; from tools to agents
To do:Assignment 2 (by next class) Watch:How to Use NotebooKLM Better than 99% of People (find other YouTube video links in the notes/slides) |
Notes | Slides |
| 5 | Feb 10 |
Using LLMs via API; GitHub Copilot and workflows; generative AI in Jupyter notebooks; custom automations
Read:Case study notes on the |
Notes | — |
| 6 | Feb 17 |
AI integration in Jupyter notebooks via the Jupyter AI extension (hands-on)
To do:Assignment 3 Watch:Real-time Collaboration Is Not Just for Humans Anymore |
notebook | — |
| – | Feb 24 |
No class — Prof. Barba at an event in Dayton, OH
Supplementary:AI Python for Beginners; Jupyter AI: AI Coding in Notebooks on Deeplearning.AI |
||
| – | Mar 3 | No class — Prof. Barba in China | ||
| 7 | Mar 17 |
AI-assisted coding: from autocomplete to agentic engineering; Copilot, Claude Code, CLI agents; MCP and skills
|
Notes | Slides |
| 8 | Mar 24 | Agentic coding practicum: building a research skill from scratch; the SKILL.md conventions |
— | |
| 9 | Mar 31 | Skills as infrastructure: evaluating, designing, and evolving agent-readable skills | Notes | Slides |
| 10 | Apr 7 |
Context engineering and agent design for research; the attention budget; context hygiene; memory
To do:Assignment 4 |
Notes | Slides |
| 11 | Apr 14 | Agentic research flow via context engineering; a compounding research codebase: testing, version control, repository structure | Notes | Slides |
| 12 | Apr 21 |
AI support for organizing and presenting research knowledge: diagramming, slide creation, note-taking, personal knowledge base
To do:Assignment 5 |
Notes | Slides |
| 13 | Apr 28 |
Organizing knowledge with AI; using AI as a tool for thought. RAG vs. the LLM-wiki pattern. Lessons for the working researcher.
Watch:How to Stop AI from Killing Your Critical Thinking | Advait Sarkar | TED |
Notes | — |
Reflections after the course, and using the materials
This course ran once, in the spring of 2026, as a special-topics seminar for engineering doctoral students. This page is the record of that semester and not a polished, evergreen curriculum. Generative AI moves quickly enough that any honest treatment of it is a time-stamp: some tools named in these notes will have changed, merged, or vanished by the time you read them, and a claim that felt current in April may already read as quaint. I left the materials as they were rather than quietly updating them, so the record stays truthful to its moment.
But the posture should outlast the particulars. The semester was less about any single product than about meeting a fast-moving technology as a "power user": someone who understands enough of how these systems work to use them deliberately, who reaches for them where they help and declines them where they don't, and who treats context engineering—deciding what an AI system sees, remembers, and is asked to do—as the skill that carries across whatever tools come next. If a specific tool dates, that habit of mind should not.
If you would like to follow that journey yourself, the schedule above is the map and the lecture notes are the main path through it. Begin where the course began, with the questions before the tools: the opening session on AI in science and the
From there the path turns practical: accelerating literature review (Class 4); moving from AI autocomplete to agentic coding across notebooks, GitHub Copilot, and Claude Code (Class 5 through Class 7); building your own agent skills and then maintaining them (Class 8, Class 9); and the two sessions where context engineering becomes explicit, including the idea that good software engineering practices—tests, version control, repository structure—are
The final sessions turn from doing to thinking: organizing knowledge, and using AI as a tool for thought rather than a replacement for it (Class 12, Class 13).
I recommend that you start with the lecture notes: they are extensive and largely self-contained; the "Read" and "Watch" links are curated supplements that extend them. Do the assignments against your own research rather than a toy problem; they were written to be domain-specific. You can expect a few dead ends. Some sessions were hands-on and live, and hard to replicate in writing; there, try to follow the linked resource and reproduce the exercise yourself.
Course Journey
