Code Mentor

Inspiration

While building projects in high school to stand out on university applications, I ran into a persistent dilemma: as someone who had only taken one computer science course at the time, my ideas were never the problem, but my lack of technical experience was. Back then, I relied on what I knew and ChatGPT to bring app ideas to life. Once I started university, I saw this exact pattern repeat. Computer science students frequently get stuck when trying to turn ambitious ideas into solid portfolio projects for co-op applications. Existing AI tools often bypass the entire learning process by generating finished code outright, leaving students to copy and paste without understanding the underlying architectural decisions or data flows.I wanted to create a solution that acts as a private tutor rather than a shortcut. CodeMentor was designed to empower students to build software independently, learn real-world architecture, and fully own their portfolio projects while keeping all local code and notes completely private on their NAS.

What it does

CodeMentor is an AI coding tutor built for UGREEN AI NAS. It scans a project folder or brief stored locally, analyzing existing notes, source code, and detected tech stacks. Once the user confirms, it generates a structured ROADMAP.md and CHECKLIST.md. Instead of generating raw code solutions, CodeMentor breaks complex features into phased learning milestones and bite-sized, actionable tasks, helping students learn by doing while building their own software.

How we built it

CodeMentor was designed using the UGREEN Skill Developer Package. The core orchestration is defined in a clean SKILL.md file that guides UGOS expert capabilities step-by-step to scan files, read documentation, and draft project plans without relying on external scripts.

Challenges we ran into

Designing a multi-step logic flow entirely upfront was a key challenge. It required sitting down and carefully planning how each capability hands off clean data to the next step, ensuring the workflow remains smooth, reliable, and easy to follow.

Accomplishments that we're proud of

We designed a concept that actively encourages deep learning, architectural understanding, and developer self-reliance rather than handing over auto-completed code solutions.

What we learned

We gained a strong understanding of UGOS agent orchestration and learned how to effectively break broad developer ideas into structured, sequential execution steps.

What's next for Code Mentor

Building the full system, and testing execution across real project repositories.

Built With

• n/a