luminafy

Inspiration

Late nights. Stack of papers. Red pen running dry. Every teacher knows the feeling. I built because grading shouldn't be a second job. I wanted to give teachers their weekends back without sacrificing personalized feedback. What if AI could handle the repetitive parts, so teachers could focus on what truly matters: connecting with students.

What it does

Upload a ZIP of assignments. Select a subject, set strictness, add custom instructions. Hit grade. AI extracts student names from files, analyzes each submission, detects AI-generated content, verifies math answers with Wolfram Alpha, and returns a CSV with scores, feedback, and plagiarism insights all in seconds. Teachers can save grading templates, manage AI rules, and track stats from a clean dashboard.

How we built it

Frontend: Next.js 14 with TypeScript, Tailwind CSS, and shadcn/ui for a responsive, modern dashboard.

Backend: Next.js API routes orchestrated by Inngest for reliable background jobs more timeouts on long grading runs.

AI & Grading: Cerebras AI powers subject-specific grading. Custom adapters handle math (with Wolfram Alpha), English, science, history, and general assignments. Each grader injects the teacher's strictness and custom rules into prompts.

Plagiarism Detection: AI analyzes submissions for AI-generated content percentage and flags high-risk work.

Database: MongoDB Atlas stores teacher profiles, preferences, custom AI rules, grading templates, and statistics.

Auth: Clerk handles secure authentication with seamless user sync to MongoDB.

File Processing: AdmZip and Mammoth extract text from DOCX files in-memory no disk writes, Vercel-friendly.

Challenges we ran into

Accomplishments that we're proud of

Teachers can now grade in minutes what used to take hours. Plagiarism detection gives educators honest insight without manual guesswork. The dashboard feels fast and intuitive teachers actually want to use

What we learned

AI grading needs human‑in‑the‑loop design: scores, feedback, and flags, not final verdicts. Serverless architectures demand different thinking buffers over files, events over synchronous jobs. Teacher workflows vary wildly; templates and custom rules make the tool adaptable. Plagiarism detection is about awareness, not punishment.

What's next for luminafy

PDF Support: Add serverless‑friendly PDF extraction. Turnitin‑style Source Matching: Integrate with plagiarism APIs. Class Analytics: Heatmaps of common mistakes, grade distributions per assignment. Student Portal: Let students see feedback and track progress. Rubric Builder: Teachers create custom rubrics, AI grades against them. LMS Integrations: Connect with Google Classroom, Canvas, Moodle.