DEV Community

exortic labs
exortic labs

Posted on

We're Building an Open-Source AI That Teaches You Instead of Giving Answers (Here's How to Help)

#ai #webdev #opensource #github

When we launched Guidey (our open-source AI tutor), the response from developers and educators was immediate:

"Finally, an AI that doesn't just blurt out the answer."

Most AI coding assistants and chatbots are designed to do the work for you. You paste a broken function, it spits out the fix, you move on.

That's great for shipping code on a deadline. It's terrible for learning.

If you want to actually understand data structures, systems design, or math, you need something that works like a human mentor: asking you what you think, pointing out edge cases, sketching diagrams, and giving you small hints when you're stuck.

We just opened up our development roadmap on GitHub, and we're looking for contributors to help build this out in the open.

What We're Building Next

Here are the big engineering challenges we're tackling right now across the stack:

1. The Interactive Blackboard (Three.js & Canvas)

Learning complex topics requires visuals. We are building an infinite interactive blackboard where:

  • The tutor can dynamically sketch graphs, memory layouts, and logic trees while explaining concepts.
  • A lightweight 3D avatar uses gaze tracking and pointing to reference specific diagrams on the board in real time.
  • Relevant issues: #10 (Infinite canvas & chalk artifacts), #13 (3D avatar inverse kinematics pointing).

2. Anti-Monologue Guardrails (Prompt & Interaction Flow)

Have you ever used an AI tutor that talks too much?
Standard LLMs love giving multi-paragraph lectures. We want Guidey to feel like a rapid, natural conversation:

  • Keep responses short and conversational (1–3 sentences max per turn).
  • Enforce strict rules so the model never accidentally spoils the solution, even when the user asks directly.
  • Relevant issue: #14 (Conversation pacing and anti-monologue guardrails).

3. Local Model Support & Modularity (Python / FastAPI)

Not everyone wants to send their learning data to cloud providers:

  • We're adding dynamic model selection so developers can run Guidey 100% locally with tools like Ollama, Llama.cpp, or vLLM.
  • Modularizing the backend into clean service layers for student sessions, history tracking, and hint evaluation.
  • Relevant issues: #5 (Custom LLM controls), #12 (Modular server architecture).

4. Clean Blackboard UI & Mobile Experience

A tutor should feel friendly, clean, and distraction-free:

  • Touch-friendly blackboard controls for tablets and mobile screens.
  • Dark chalkboard aesthetic with crisp typography and responsive layouts.
  • Relevant issue: #11 (Blackboard viewport layout & typography overhaul).

How to Get Started

You don't need to be an AI researcher to contribute. We have issues across frontend design, Python backend, 3D graphics, prompt tuning, and documentation.

  1. Check out the live app: guidey.exortic.com
  2. Explore the issues on GitHub: github.com/exorticlabs/guidey/issues
  3. Run it locally: 
   git clone https://github.com/exorticlabs/guidey.git
   cd guidey
   # Install dependencies and start the dev server
Enter fullscreen mode Exit fullscreen mode

Pick an open issue, leave a comment if you'd like to work on it, and let's build an AI that actually teaches.

Top comments (0)