Assignment Autopsy

Inspiration

Assignment Autopsy was built based off what currently almost all educational, assignment-focused apps like Microsoft Teams (with its "Tasks" app) and Google Classrooms lacked: proper guidance on students in their tasks.

What it does

Assignment Autopsy resumes to the following:

Teacher creates an assignment
            │
            ▼
     Requirements + Rubric
            │
            ▼
       Student works
            │
            ▼
       Student uploads
            │
            ▼
      AI pre-submission review
            │
       ┌────┴─────┐
       ▼          ▼
    Fix work   Submit anyway
       │          │
       └────┬─────┘
            ▼
       Final submission
            │
            ▼
     Teacher reviews work
            │
            ▼
      Official evaluation

Unlike other apps, the AI compares the student's submission with information provided by the teacher.

It can identify things such as:

• requirements that appear to be missing
• requirements that may be incomplete
• possible issues with the assignment
• how the work appears to match each rubric criterion
• evidence supporting its assessment
• situations where there is not enough information to make a confident assessment

The AI does NOT decide the student's final grade.

How we built it

This was built using Python 3 and libraries like FastAPI, Jinja2, SQLAlchemy, Alembic, msgspec, etcetera.

The source code is fully scalable, as it is divided by horizontal functionality layers, allowing each part to be easily identified and updated.

Challenges we ran into

The main challenges in the project were the frontend UX and how the AI would treat the students info.

The frontend had to be easy-to-use for everyone, and keeping a familiar UI from other apps applicable here.

Another big challenge was how to safely guard the AI for simply returning any answer or getting its instructions deleted by a student's data input.

Accomplishments that we're proud of

We're really proud on how it all ended up looking, not just UI & UX wise in the fronted, but the backed. It is really clean and has everything divided correctly and does exactly what it suggests, meaning that when we were bug catching and testing, whenever something failed it pinpointed right where it was from, without having to look at a massive thousand lines file.

What we learned

Assignment Autopsy had to be secure for all, this means that, as a bare minimum, accounts should have a safer option to secure their accounts other than email and password. That is why we designed a TOTP/2FA protocol and allowed Google OAuth2 Sign-In whenever possible.

What's next for Assignment Autopsy

We are really looking forward on improving and adding new features to the application, but as it looks right now, it has everything that should be needed for it to actually be used in any educational center.

AI Disclosure

AI has been used to build the frontend, and for brainstorming ideas for this hackathon.

Built With

• alembic
• argon2
• fastapi
• featherlessai
• httpx
• jinja
• msgspec
• postgresql
• python
• sqlalchemy