Pathwise

Inspiration

Most classrooms still run at one speed. Thirty kids get the same lesson, the same explanation, and the same worksheet, and anyone who learns differently either gets bored or gets left behind. Teachers know this. They just don't have the hours to write thirty versions of every lesson.

What it does

In the first week, each kid does a learning-style check-in: do they get it from pictures, from hearing it, or from doing it with their hands? Then every kid gets an AI agent that remembers this and keeps notes all year. When the teacher sets the day's lesson, like long division, a teacher agent sets the goal and each kid's agent writes them their own plan to get there. Same destination, different road for every kid.

How we built it

It's a small Python server and a plain web page. The agents run on Qwen3, an open-source model, through Ollama on a laptop. Each kid's profile and memory are saved to a file, so the notes carry over from one lesson to the next.

Challenges we ran into

Our first version just showed plans we wrote by hand, which wasn't the point, so we rebuilt it to call a real model. We also had to check the model's math, because it doesn't always get division right. And running six agents one after another on a laptop is slow. Our first full run timed out before the teacher agent had finished the lesson goal.

Accomplishments that we're proud of

The agents can't change the lesson. They can only change how each kid gets there. One kid's plan can even hold back the hardest problem if that kid's notes say they aren't ready for it yet.

What we learned

Personalizing a lesson isn't about swapping in different content. It's about pacing, format, and knowing where each kid usually gets stuck.

What's next for Pathwise

Next up are faster models so a full class finishes in minutes, a real intake test, the jump from five kids to a full class of 30, and a dashboard where teachers can see and edit every plan.

Built With

• ai