Momentum

Background & Motivation

Teachers spend hours collecting, reading, and grading exams-especially in larger classes or when work is handwritten. Submissions get lost, feedback is delayed, and bottlenecks pile up. Momentum was built to remove these barriers, letting teachers create and manage exam pages, collect scanned or written work, and review AI-assisted grading, all in one place.

Solution

Each teacher creates an exam, sets questions, and invites students by email. Students upload scanned answer sheets or essays; the system accepts any common format. All files go directly to encrypted storage-no manual handling, no email threads.

Once an exam is uploaded, a Supabase Edge Function triggers an AI task: GPT-4 Vision parses each submission, extracts answers, and maps them to the right questions. A second AI review task grades the answers and prepares a report. The teacher can accept, adjust, or override grades from a dashboard. Every change is tracked.

Technical Architecture

• Frontend: Next.js SPA for both teachers and students.
• Backend: All core logic is handled via Supabase serverless (Edge Functions).
• AI Extraction: OpenAI GPT-4 Vision (no custom OCR).
• Auth & Security: Supabase Auth with Postgres RLS for both teachers and students. No user can access or even list resources they don’t own.
• Data Storage: Files encrypted at rest, strict row-level security for all metadata and content.

Impact

• Average grading time per exam reduced from 8–10 minutes to under 3 minutes, even for handwritten work.
• Upload failures dropped below 1% after onboarding improvements.
• Teachers reported faster feedback cycles, less administrative hassle, and fewer missing submissions.
• All data handling passed GDPR audits in live pilots.

Failures & Lessons

• Some scanned documents were unreadable by GPT-4 Vision (blur, shadows). The system allows manual review and direct teacher override for any failed extraction.
• Edge Function concurrency sometimes caused delays under heavy load; mitigated by batching and status notifications to teachers.
• A minority of teachers preferred manual grading for subjective work-Momentum allows disabling automation per exam.

Summary

Momentum does not promise “AI replaces grading.” Instead, it automates file collection, extraction, and initial review-letting teachers reclaim time and control. Security, privacy, and teacher oversight are never optional. Adoption spreads because it works with real classroom habits and does not force new bureaucracy.