Gradr

• 

  Part 1 completion mail

• 

  Part 2 completion mail

• 

  Chatbot completion and test

• 

  Gradr assessment page

• 

  Grading completed page

• 

  Grading details page

Inspiration

During my MSc., I had just begun to learn to code so I decided to use my newly developed skills to solve a problem that I had observed within my environment - manual grading. This process usually involves a lecturer using a guide to assess and score student's answers to a quiz.

What it does

The user uploads a test question (Microsoft word document), a marking guide (Microsoft word document) and students' answers (images). The application then performs grading and scores the students appropriately.

How we built it

The application's process can be broken down into 3 simple steps

• Text extraction and Embeddings generation - the text from the question and the marking guide are extracted and stored them along with their embeddings in Azure CosmosDB for MongoDB.
• Prompting - we pass the student's answers as a prompt to the ai
• Grading - the ai grades the student's attempt based on the data in the DB.

Challenges we ran into

• I encountered quite a number of challenges during the training phase. From deployment using the bicep file to hosting/deploying the UI and backend api.

Accomplishments that we're proud of

• We were able to rebuild our project (originally built with Google Gemini) using the Microsoft Copilot stack.

What we learned

• The RAG process

What's next for Gradr

• Handle other forms of student's answers e.g PDFs, GitHub links etc
• Test with real data.
• Get our first set of paying customers.

Built With

• azure-file-share
• azurecosmosdb
• azureopenai
• javascript
• node.js