Skip to content

AI for EE Research

Title: AI for EE Research: From Researcher to Orchestrator

Target Audience: Graduate students (MS and PhD) in electrical engineering, computer engineering, and computer design, plus postdocs and faculty who want practical AI skills for everyday research

Prerequisites:

  • Graduate standing (or equivalent research experience) in electrical engineering, computer engineering, or a related field
  • Working knowledge of at least one programming language, preferably Python
  • Basic comfort with a command line, a text editor, and Git
  • Experience writing a technical paper or report
  • No prior machine learning, prompt engineering, or agent experience required

Course Overview

Graduate research in electrical engineering and computer design is a pipeline of repeatable tasks: finding and reading literature, brainstorming and scoping a problem, writing analysis and simulation code, processing data, and preparing figures and manuscripts. Generative AI tools and agents can now take a substantial share of each of these tasks, but only for researchers who know how to direct them, check their output, and take responsibility for the results.

This course is a hands-on guide to that skill. Its central idea is a change in the researcher's role: from doing every task personally to orchestrating a team of agents. In the vision developed throughout the book, each researcher coordinates roughly 20 planning agents, which explore literature, decompose problems, and draft specifications, and 10 execution agents, which write code, run analyses, build simulations, and prepare figures and manuscripts. The researcher's job is to set direction, supply domain judgment, resolve conflicts between agents, and decide what is true. The researcher stays accountable for the science.

The course is practical rather than theoretical. Students learn the two building blocks that make agents reliable in a research setting: plain Markdown files that capture a project's goals, conventions, and constraints, and reusable agent skills that turn a repeatable research task into a single, testable command. Throughout, the course works from case studies of real work, including AI-generated interactive simulations (MicroSims), intelligent textbooks, and a full paper prepared for arXiv with AI agents, and it treats effective, creative, responsible, and ethical use as equal goals.

The course also teaches personal knowledge graphs: a researcher's own notes, papers, experiments, people, and projects, represented as linked entities and relationships rather than as a folder of files. A personal knowledge graph gives agents a structured, queryable memory of the researcher's work, and the course shows how to build one with an agent, query it, use it to ground a model's answers, and decide what an agent may be allowed to see.

The course also surveys the new class of always-on personal agents that arrived in 2026, because researchers will meet them in their own labs whether or not a course introduces them. These are hosted or self-hosted agents that keep their own cloud computer, sign in to tools on the user's behalf, remember context across sessions, and work while the researcher is away. As of this writing the survey covers SpaceXAI's Grok Bot, OpenAI's Dots, Meta's Muse, Anthropic's Claude Cowork, and the open-source OpenClaw and Hermes Agent. The goal is not to rank products, which change monthly, but to give researchers durable criteria for choosing and supervising any of them: where the agent runs, what it can reach, how permissions and approvals work, what it costs, and what happens to research data. The current research notes are collected in the appendix Personal Agent Software.

Main Topics Covered

  1. The research workflow and where AI helps: literature search, data analysis, brainstorming, coding, and manuscript preparation
  2. How large language models work, at the level a researcher needs: tokens, context windows, hallucination, model selection, and cost
  3. Prompting and context engineering for research tasks
  4. Markdown as a project knowledge format: project instruction files, notes, specifications, and conventions that agents read before they work
  5. Agent skills: designing, writing, testing, and sharing reusable skills for repeatable research tasks
  6. Agents and tool use: file access, shell commands, web search, subagents, hooks, and connecting to external tools and data
  7. Planning agents and execution agents: dividing work, handoffs, and coordination
  8. Orchestration: the researcher as the coordinator of 20 planning agents and 10 execution agents
  9. AI-assisted literature search, reading, and synthesis
  10. AI-assisted data analysis and scientific coding, including scripts for simulation, signal processing, and plotting
  11. AI-assisted brainstorming, problem framing, and hypothesis development
  12. AI-assisted manuscript preparation: structure, LaTeX, figures, references, and arXiv submission
  13. Interactive simulations (MicroSims) and intelligent textbooks as ways to communicate and teach research results
  14. Verification, validation, and reproducibility of AI-assisted work
  15. Responsible and ethical use: citation, disclosure, privacy, intellectual property, bias, and academic integrity
  16. Case studies of AI-assisted research and publication workflows, including the MicroSims arXiv paper and intelligent textbook projects
  17. Personal agents: always-on hosted and self-hosted agents such as Grok Bot, Dots, Muse, Claude Cowork, OpenClaw, and Hermes Agent, including their architectures, permission and approval models, multi-agent features, pricing, and data-handling trade-offs
  18. Personal knowledge graphs: representing a researcher's notes, papers, and projects as entities and relationships, building them with agents, querying them, using them as grounded agent memory, and protecting their privacy

Topics Not Covered

  • Training, fine-tuning, or building large language models from scratch
  • Machine learning theory, deep learning architectures, and neural network mathematics
  • AI hardware design, accelerators, and chip architecture for AI workloads
  • Vendor-specific electronic design automation (EDA) flows and tool tutorials
  • Domain-specific EE theory such as circuit analysis, device physics, or quantum electronics (the course assumes students bring this knowledge)
  • Legal advice on copyright, patent, and data-protection law
  • Comparative benchmarking and rankings of commercial AI products
  • Using AI to replace the researcher's judgment, authorship, or accountability

Learning Outcomes

After completing this course, students will be able to:

Remember

Retrieving, recognizing, and recalling relevant knowledge from long-term memory.

  • Define the key terms: token, context window, prompt, agent, subagent, skill, tool call, hallucination, and orchestration
  • List the five everyday research tasks this course targets: literature search, data analysis, brainstorming, coding, and manuscript preparation
  • Recall the purpose of a project instruction file and the typical sections it contains
  • Identify the parts of an agent skill: name, description, instructions, and supporting files
  • Name common failure modes of AI output, including fabricated citations, silent errors in code, and overconfident summaries
  • State the difference between a planning agent and an execution agent
  • Name the major personal agent products (Grok Bot, Dots, Muse, Claude Cowork, OpenClaw, and Hermes Agent) and the vendor or project behind each

Understand

Constructing meaning from instructional messages, including oral, written, and graphic communication.

  • Explain why a language model can produce fluent but false statements
  • Describe how context window limits affect long research tasks
  • Explain why Markdown is a good format for knowledge that both humans and agents read
  • Summarize how an agent differs from a chatbot, and how tool use extends what it can do
  • Describe how an always-on personal agent differs from a session-based assistant, including where it runs, how it signs in to tools, and how it keeps memory between sessions
  • Explain how a personal knowledge graph differs from a folder of notes, and how entities, relationships, and backlinks make a researcher's knowledge queryable by an agent
  • Explain how splitting work between planning and execution agents reduces errors and cost
  • Describe the researcher's responsibilities when AI contributes to a result, including citation and disclosure

Apply

Carrying out or using a procedure in a given situation.

  • Write a project instruction file that gives an agent the goals, conventions, and constraints of a research project
  • Use an AI tool to run a literature search and produce an annotated reading list with verified citations
  • Prompt an agent to write, run, and debug an analysis script on a real dataset
  • Create a reusable agent skill that automates one repeatable task from their own research workflow
  • Use an agent to draft a manuscript section from notes, then revise it into their own voice
  • Generate an interactive simulation that illustrates a concept from their research
  • Build a personal knowledge graph from their own research notes and papers using an agent for entity extraction, then query it to ground an agent's answers

Analyze

Breaking material into constituent parts and determining how the parts relate to one another and to an overall structure or purpose.

  • Decompose a research project into tasks suited to planning agents and tasks suited to execution agents
  • Compare the output of several agents on the same task and identify where and why they disagree
  • Trace an error in AI-generated code or a claim in AI-generated text back to its source
  • Distinguish tasks that benefit from AI assistance from tasks that require the researcher's direct work
  • Analyze a case study of an AI-assisted research project and map its workflow, handoffs, and review points
  • Compare personal agents on architecture, permission and approval model, multi-agent support, integrations, and cost, separating vendor claims from independently tested behavior

Evaluate

Making judgments based on criteria and standards through checking and critiquing.

  • Verify AI-generated citations, numerical results, and code against primary sources and independent tests
  • Judge whether an AI-assisted result meets the standard for reproducibility in their field
  • Assess a skill or agent workflow for reliability, cost, and risk before relying on it
  • Critique an AI-assisted manuscript for accuracy, originality, and proper attribution
  • Decide when and how to disclose AI assistance to a venue, advisor, or collaborator under ethical and institutional guidelines
  • Assess the security and data-handling risks of giving a personal agent access to research accounts, unpublished data, and licensed tools before adopting it
  • Decide which parts of a personal knowledge graph an agent may read, and assess the privacy risk of exposing unpublished results through it

Create

Putting elements together to form a coherent or functional whole; reorganizing elements into a new pattern or structure.

  • Design an orchestrated team of planning and execution agents for a research project, with defined roles, handoffs, and review checkpoints
  • Build a personal library of agent skills and project instruction files for their research group
  • Produce a complete, verified research artifact with agent assistance, such as a literature review, an analysis notebook, or a draft paper
  • Develop a responsible-use policy for AI in their lab or research group
  • Capstone project: plan and run an end-to-end research task with a coordinated team of agents, document the workflow, and present the verified results along with an honest account of where human judgment was required