-
Notifications
You must be signed in to change notification settings - Fork 2
Applied Example Microelectronics Tutor Demo
type: synthesis up: "Home_llm-wiki-memory-template" tags: [application, demo, education]
A conference demonstration of the wiki-llm pattern applied to undergraduate microelectronics education, using the Purdue SCALE curriculum (Arduino UNO + Elegoo Super Starter Kit). Treated as an empirical test of the template in a non-research domain.
A derived project (microelectronics-tutor-demo) instantiated from this template hosts a small wiki covering roughly 15 concept pages across three clusters:
-
Components:
Arduino-Uno-Board,LED-Basics,RGB-LED,Current-Limiting-Resistor,Pushbutton-Switch, ... -
Concepts:
Forward-Voltage,Pulse-Width-Modulation,Common-Anode-vs-Common-Cathode,Floating-Input-and-Pull-Up-Resistors, ... -
Code patterns:
Arduino-Sketch-Structure,Blink-Pattern,analogWrite-for-PWM, ...
Typed edges between pages encode the curriculum's prerequisite graph: RGB-LED --extends--> LED-Basics, LED-Basics --requires--> Current-Limiting-Resistor, RGB-LED --uses--> Pulse-Width-Modulation, and so on. The graph is what makes the agent's Socratic navigation through prerequisites possible.
Two short videos, same student, same Lab 1 assignment:
- Video 1 (the shortcut, ~90 seconds). Student pastes the assignment into a generic Claude or ChatGPT chat, copies the response, submits. A follow-up diagnostic question reveals they cannot reason through the underlying concept.
- Video 2 (the tutor, ~3 to 4 minutes). Same starting question, but the student is in a Claude Code session against the course wiki. The agent soft-redirects, scaffolds through the wiki pages, asks Socratic questions, watches the student reason through to their own answer.
The diagnostic question, "if you'd used a blue LED instead of a red one, would 220Ω still be the right resistor?", lands in both videos. The Video 1 student hesitates or guesses; the Video 2 student explains forward-voltage and recalculates.
-
Three operations under domain shift. Can
/wiki-sourceingest curriculum PDFs cleanly? Does/wiki-experimentadapt to "filed lesson outcomes" instead of "research findings"? The operations are domain-agnostic on paper; this is an empirical test of that claim. -
Typed-edge traversal at the user-facing level. The agent's navigation across
RGB-LED → Common-Anode-vs-Common-Cathode → Pulse-Width-Modulationis the visible demonstration of Typed-Edges-and-Frontmatter. The student sees the graph being used. - Soft-redirect behavior. The pattern this demo encodes is "AI that knows the course, not a gatekeeper": the agent always offers a constructive path, never refuses to help, and gracefully falls back to web search for out-of-scope questions.
Conference talk scheduled for 2026-05-19. Wiki authoring and video scripts staged at microelectronics-tutor-demo/docs/demo/. Production work (filling the ~15 concept pages, recording the two videos, building the framing/closing slides) is the next phase.
- Not an anti-cheat tool; no detection or enforcement.
- Not an LMS integration; standalone tool that runs alongside the course.
- Does not track student progress for grading; per-student memory is opt-in and stays local.
The pitch is AI that knows the course makes exploring the course material the easy path, not an obstacle to it.