GhostGuide — An AI Coding Mentor That Guides, Not Spoils

Inspiration

AI coding assistants can generate solutions in seconds, but finishing a task is not the same as understanding it. For students learning programming, immediately receiving a complete solution can reduce the opportunity to practice independent problem-solving.

GhostGuide explores a different approach: What if an AI assistant behaved more like a mentor than an answer machine?

Inspired by the concepts of cognitive scaffolding and metacognition, GhostGuide aims to help learners solve difficult programming problems through guided reasoning, incremental hints, and reflection.

The Halloween theme gives this learning experience a playful twist: learners explore challenging coding puzzles alongside a friendly ghost that reveals clues instead of immediately revealing answers.

What It Does

GhostGuide is designed around five core experiences:

• Progressive Hint System: Instead of revealing an entire solution, the AI provides increasingly specific hints as learners need them.
• Socratic Debugging: The assistant asks targeted questions to help users identify errors in their own reasoning and code.
• Adaptive Guidance: Support can become more specific when a learner is stuck, encouraging independence whenever possible.
• Learning Reflection: After completing a challenge, users can review their reasoning, mistakes, and key takeaways.
• Haunted Learning Experience: A Halloween-inspired interface turns debugging and problem-solving into an interactive adventure.

The ultimate goal is to help users become better independent problem-solvers, not simply faster consumers of AI-generated code.

How We Built It

The implementation plan uses Kiro as the AI-assisted development environment, with an emphasis on spec-driven development.

The planned workflow separates the application into its user interface, learning session management, guided hint generation, and progress tracking.

Kiro specifications can help define requirements and break implementation into manageable tasks, while steering documents can maintain consistent architectural and coding conventions.

Implementation details to update before final submission: Document the actual framework, AI model or API, backend, storage, Kiro features, and tests used in the working application.

Challenges

The central design challenge is balancing helpful guidance with independent learning.

An assistant that reveals too much can undermine the purpose of the experience. However, an assistant that provides vague hints may frustrate users.

The proposed approach is a staged hint system that moves from conceptual guidance to increasingly concrete assistance.

Other important engineering challenges include maintaining conversational context, preventing premature solution disclosure, and creating a smooth learning experience.

What We Learned

This project is an opportunity to investigate how AI assistance can support genuine skill development rather than only improving short-term task completion.

The development process is also intended to explore how structured requirements, AI-assisted implementation, testing, and iteration can work together in Kiro.

To complete after development: Add specific technical lessons, unexpected problems, tests performed, and how those problems were resolved.

What's Next

Future possibilities include personalized learning paths, algorithm visualization, programming contest practice, and more detailed progress analytics.

The long-term vision is an AI learning environment in which the most valuable outcome is not the code generated, but the understanding gained by the learner.

GhostGuide: The best hint is the one that helps you discover the answer yourself.

Built With

• kiro
• node.js
• react
• tailwind-css
• typescript
• vite