Cham

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Cham

The problem

Beginners now ask AI tools for code and get more than they can understand. Research backs the worry. In a field experiment with about 1,000 high school students, unrestricted GPT-4 help raised practice scores by 48% but lowered later exam scores by 17% (Bastani et al., PNAS 2025). Sweller's cognitive load theory explains part of it: a novice who spends working memory on tooling has less left for learning the concept.

Our solution

Cham is a single Rust binary that runs a tiny route language:

when someone posts /orders:
    need name, phone, item
    time = now
    order = save name, phone, item, time to orders
    answer "Thanks, " + name, order

Each line is one step, run in order. Storage adapts to new fields by itself. When something breaks, Cham names the line or field and suggests a fix in English, Ukrainian or Russian, instead of printing a stack trace.

Demo apps

1. Order page for a local shawarma stand with a live kitchen queue.
2. Equipment booking for a university lab that rejects double bookings.

Where AI fits

Cham does not fight AI tools. It gives them a smaller target. A model can write a .cham file, and the student can still check every line of it. The README cites the studies (Sweller 1988; Bastani et al. 2025; Lee et al., CHI 2025; Becker et al., METR 2025) with links.

Built With

• css3
• html5
• javascript
• rust
• serde-json
• tiny-http