Verity Assessment Studio

Inspiration

Teachers spend valuable time marking learner scripts question by question. We wanted to explore whether AI could reduce this repetitive work without taking professional judgement away from educators.

That became Verity Assessment Studio.

AI proposes. Teacher decides.

What it does

Teachers upload a memorandum and learner script. Verity evaluates each answer and proposes a mark, explanation and confidence score.

When Verity is uncertain, it flags Teacher Review Needed instead of guessing. The teacher confirms or changes the mark, and only the teacher can sign off the final result.

How we built it

Verity runs locally using React, Vite, TypeScript and Tailwind CSS for the frontend, with FastAPI, SQLAlchemy and SQLite on the backend.

A local Ollama Qwen2.5 model performs the AI-assisted marking. TXT and text-based PDF files are supported.

Challenges we ran into

Our biggest challenge was scope. Verity could include handwriting OCR, bulk scanning, QR separation, analytics and school integrations, but building everything would hide the core idea.

We focused on one powerful experience:

Memo → Script → AI proposes → Teacher reviews → Teacher signs off.

AI reliability was another challenge, so uncertainty is deliberately escalated to the teacher instead of being hidden.

Accomplishments that we're proud of

We're proud that the educator remains in control.

Verity doesn't just generate a score. It explains its proposed mark and identifies answers that need human judgement.

Most importantly, no mark becomes final without educator sign-off.

What we learned

We learned that responsible AI isn't only about making AI more capable. It's also about knowing when AI should stop and let a human decide.

The goal isn't autonomous marking. It's AI-assisted marking with human authority.

What's next for Verity Assessment Studio

Next, Verity could expand to handwriting OCR, bulk scanner processing, multiple learner scripts, class analytics and school-system integration.

But one principle will remain:

AI can propose a mark. Only the educator can make it final.

Built With

• api
• css
• fastapi
• ollama
• pdf
• python
• qwen2.5
• react
• rest
• sqlalchemy
• sqlite
• tailwind
• typescript
• vite