Effortless Assessments: Automated Grading & Cheat Beating

To save teachers time and stress, and accelerate students' learning, we're adding open-response question grading, feedback and 'cheat detection' to our AI-powered Quiz tool for teachers, Conker.ai.

Comment

Inspiration

We're an Edtech company and we love solving problems for educators and students. Teachers spend 10% of their time grading and giving feedback, but still fail to meet students' needs for timely and effective feedback on their work.

What it does

Our Conker.ai platform already offers teachers the ability to generate new quizzes and assessments using AI, but these have been limited to 'closed' response questions that have right/wrong answers.

For our hackathon project, we are adding the functionality to generate two kinds of open-response question from any text stimulus provided by a teacher. These question types feature on the online STAAR testing in Texas, which is where we are focusing our efforts and working with pilot school districts.

We are also building a simple 'cheat detection' classifier model, based on keystroke analysis of user input.

How we built it

  • Generating questions: OpenAI Completion API on GPT 4 model
  • Grading questions: OpenAI Completion API on GPT 3.5 model
  • Front End UX: extending next.js codebase hosted on Vercel, applying new question types, auto-grading behaviour and cheat detection data labelling UX
  • Cheat detection: Fast.ai 3 layer classification model with training data collection pipeline

Challenges we ran into

Generating consistently good questions, accurate grades and high quality feedback all posed challenges for prompt engineering. Defining and training a cheating detection model with a cold-start labelling data collection pipeline

Accomplishments that we're proud of

Coming up with a new product concept (cheating detection using keystroke analysis) which addresses a growing pain point for schools, and fits with our strategy of simplifying assessment.

What's next for Effortless Assessments: Automated Grading & Cheat Beating

  • Test with users - we hope to launch this next week
  • Extend automated grading with more question types, e.g. AP-style essay grading
  • Extend "Cheat detection" with a data-feedback loop and potentially other signals of authenticity.

Built With

Share this project:

Updates

Submission history