mntr.ai Educational Platform

Inspiration:

Our inspiration was our teammate Nate, a one on one Computer Science tutor, that has a rough time identify the gaps in his students knowledge. With the personalized nature of one on one tutoring, we believed this is the ideal solution for AI to assist keeping track on his students progress.

What it does:

mntr.ai is a one on one tutoring platform where the tutors can assign asynchronous lectures before the tutoring session to introduce the students to the material before their sessions. The student is given a place to take notes while watching the video. Upon submission, we use AWS Transcribe to create a transcription from the uploaded lecture and Claude 4.5 Sonnet from AWS Bedrock compares the written notes with the transcript from the assigned video. Claude will automatically identify gaps in the students notes in a concise summary at the top of their submission, giving instructors quick and valuable insights to understand what misconceptions their students don't have and the power to take action.

How we built it:

We use a FastAPI backend connecting to a Redis store for in memory SocketIO connections and SQLite (which can be easily migrated to SQL or an SQL compatible database for scale) for long term storage of assignments.

Used AWS services include AWS S3 for storing uploaded files, AWS EC2 for hosting, and AWS Transcribe/Bedrock for our intelligent systems.

Our frontend is built entirely using HTML/CSS/JS without a frontend framework. While unconventional, this makes the experience responsive irrespective of machine performance, which is especially critical in educational settings.

Challenges:

One of our largest challenges was ensuring efficient parallelization of work. While this was easy at first, towards the end of the night we found it hard for multiple people to work on the project at once while we finally implemented the REST API and SocketIO communication between the frontend and backend.

We also had minor issues attempting to prompt Claude to behave in a way we expected and one that was not redundant to the core purpose. Claude was predisposed to write long, which subverted the desire of a system that could automatically summarize across two modalities of data. Overall, we somewhat remedied this issue by using a two pass design, but would've liked to spend more time perfecting it's outputs.

What's next for mntr.ai Educational Platform?

We believe that we've barely even scratched the surface of what's possible with mntr.ai. Practical features we'd be interested to include are multiple assignment types and better communication features between student and instructor.

In addition, we'd like to expand our use of in app intelligence through AWS bedrock by offering more features to give instructors insights, like performing an automatic quiz after taking notes to see what the student has committed to memory and what they only have notes on. This could possibly be followed up with dynamic flashcards to help the student review.

Despite our currently display of mntr.ai being relatively small, we believe that its resulting scope could be immense.

Built With

• ai
• api
• awsbedrock
• css
• ec2
• fastapi
• gptoss120
• html
• javascript
• python