Lab / research-notes
A teacher with no CS background building local AI tools: a field report
First-person account of using conversational AI (vibe coding) to design and ship educational software — what works, what the real limits are, and why local-first matters for classroom privacy.
- Type:Research note
- Status:In progress
- Updated:May 25, 2026
- Resources:0
- Apps:0
Section 1
Context
Most discourse around AI in education focuses on teachers using AI tools with students. This note documents a different use case: a classroom teacher using AI as a construction partner to build the educational software he then deploys with his students. No cloud APIs in the classroom, no student data sent to external servers.
Section 2
Hypothesis
A teacher with deep pedagogical knowledge but no traditional software engineering background can use conversational AI (vibe coding) to ship functional, privacy-preserving educational tools — if the pedagogical specification is precise enough. Domain expertise replaces technical expertise as the primary input.
Section 3
Implementation
Over two years, Luis Vilela Acuña built nine educational web applications using this method. Each tool was born from a specific classroom friction, built through iterative conversation with AI models, tested in real classroom sessions, and revised based on what failed. None required hiring a developer. All are free, open source (GPLv3 goal), and privacy-first by design.
Evidence note in classroom
The method works — with caveats. Tools built this way fit their original purpose with unusual precision because the specification comes from lived classroom experience. The real cost is iteration time, which remains high. Whether it is transferable to other teachers without support is an open question.
Evidence prompt: What classroom evidence confirms this outcome, and what evidence challenges it?
Action for the next cycle
Document prompt templates by pedagogical need category. Run a structured pilot with two other teachers to test whether the method is replicable. Publish a structured report on the privacy architecture decisions made across the EDUmind tool suite.
Open question: What is the smallest next experiment that could validate this iteration?
Reusable resources
Resources connected to this lab entry.
No resources linked yet.
Related learning paths
Sequences where this evidence can be reused.
No learning paths linked yet.
Related apps
Tools involved in this experimentation cycle.
No apps linked yet.