Problem Statement
Coding platforms usually keep practice, grading, and assessment separate. Students can memorize fixed question banks or trust AI-generated answers that were never validated, and assessments offer little proof that the work was genuinely the learner's own.
Proposed Solution
NeuroCode brings adaptive learning, execution-verified problems, and integrity-aware assessment into one workflow. AI-generated problems are validated by running reference solutions in a sandbox before students see them, submissions are graded against real test cases, and successful learners receive credentials that anyone can verify online.
Key Features
- Adaptive roadmap that reorders learning topics based on the learner's weak areas.
- AI-generated coding problems, validated by executing reference solutions in a Piston sandbox before being saved.
- Submission grading with real test cases and Tree-sitter structural code analysis.
- Proctored assessments with server-scored integrity signals: tab switching, large pastes, camera checks, and typing rhythm.
- Credential issuance with a public verification page, QR code, and PDF export.
- Learner dashboard, profile, leaderboard, and mock interviews, plus an educator and admin panel.
System Architecture Flow
Engineering Challenges
- AI-generated problem quality was a risk, so every generated reference solution is executed in a sandbox before the task is saved.
- The Piston runtime container could start with no language runtimes loaded, which was fixed by recreating the container and documenting the setup steps.
- Phase-wise testing exposed an assessment navigation bug that caused a blank screen and camera permission edge cases, which were fixed and re-tested.
Lessons Learned
- Owning a large system alone, from architecture and database design to AI integration, testing, and deployment.
- Designing a learning loop where generation, validation, and grading all happen before a student sees a problem.
- Balancing AI assistance with safe learning constraints, keeping feedback hint-based instead of leaking full answers.
Results
Built end to end as a solo project across different development phases, with phase-wise testing reports, a setup and deployment guide, and a user guide.