I built an AI that generates lesson plans in your exact teaching voice (open source)

Dev.to / 3/24/2026

💬 OpinionDeveloper Stack & InfrastructureTools & Practical UsageModels & Research

Key Points

  • EDUagent is an open-source tool that generates new lesson plans, worksheets, assessments, and slide deck outlines by learning a teacher’s existing materials and teaching “voice.”
  • The workflow lets users ingest a folder of prior lesson plans, extract patterns like teaching style, structure, vocabulary level, and assessment preferences, and then generate multi-week units and daily plans with those constraints.
  • It supports local execution (e.g., via Ollama for free use) or API-based model choices (examples include Claude and GPT-4o) depending on the configured backend.
  • The project emphasizes privacy by enabling local runs where the user’s materials stay on their machine, and includes differentiation guidance for struggling/advanced/ELL students.

The problem every teacher knows

You've spent years building your curriculum. Lesson plans, worksheets, slide decks, assessments — thousands of hours of work sitting in a Google Drive folder or on a flash drive.

Every time you need to create something new, you start from scratch. Or worse, you copy-paste from old files and spend an hour making it fit.

What if your old materials could become an AI that generates new ones in your exact voice?

What I built

EDUagent — open-source, runs locally or via API.

You point it at a folder of your existing lesson plans. It reads them, learns your teaching style, your vocabulary level, your structural preferences (do you always use exit tickets? graphic organizers? I Do / We Do / You Do?), and your assessment approach.

Then you ask it:

eduagent full "Photosynthesis" --grade 8 --subject science --weeks 3

And it generates:

  • Complete 3-week unit plan with essential questions and enduring understandings
  • Daily lesson plans written in YOUR voice (not generic ChatGPT voice)
  • Student worksheets, ready to print
  • Assessments, rubrics, slide deck outlines
  • Differentiation notes for struggling/advanced/ELL students

How persona extraction works

The key insight: teachers have a distinctive voice. If you've written 50 lesson plans, an LLM can learn:

  • Teaching style: Socratic? Direct instruction? Inquiry-based?
  • Structural preferences: Do you always start with a warm-up? End with an exit ticket?
  • Vocabulary level: How do you talk to your students?
  • Assessment style: Multiple choice? Rubric-based? Portfolio?

EDUagent extracts these patterns and uses them as constraints when generating new content. The output sounds like you wrote it — because it learned from you.

Runs locally (free) or via API

eduagent config set-model ollama   # Free, runs locally with Ollama
eduagent config set-model anthropic  # Best quality, uses Claude
eduagent config set-model openai     # Uses GPT-4o

With Ollama, it's completely free. Your materials never leave your machine.

Quick start

pip install eduagent

# Point it at your lesson plans
eduagent ingest ~/Documents/my-lesson-plans/

# See what it learned about you
eduagent persona show

# Generate a full unit
eduagent full "Cell Division" --grade 10 --subject biology --weeks 2

Open source, MIT license

Repo: https://github.com/SirhanMacx/eduagent

Built with a K-12 teacher as the primary user. The goal isn't to replace teachers — it's to eliminate the administrative grind so teachers can spend more time actually teaching.

Contributions welcome — especially from teachers who want to test it on their actual materials.