SupaTutor

Inspiration

We kept coming back to a problem that almost every student has experienced: you can spend hours stuck on one line of code, one math step, or one paragraph without knowing what to try next. AI can give you the answer in seconds, but getting the answer is not the same as understanding it. We wanted to build something that gives students the kind of support they might receive from a patient tutor—someone who looks at their work, finds the point where they became confused, and helps them move forward without taking over the assignment.

What it does

SupaTutor is a learning assistant for college students working on coding, math, and writing assignments. Students add an assignment to the SupaTutor workspace along with what they have tried, any errors they received, and useful notes or course materials. They can then talk to the SupaTutor agent through Superhuman Go. Instead of producing a complete solution, the agent reads the student’s actual work, asks focused questions, explains unfamiliar concepts, and gives one useful hint at a time. It also saves attempts and feedback so students can return to previous explanations and recognize ideas they may need to practice.

How we built it

SupaTutor is a learning assistant for college students working on coding, math, and writing assignments. Students add an assignment to the SupaTutor workspace along with what they have tried, any errors they received, and useful notes or course materials. They can then talk to the SupaTutor agent through Superhuman Go. Instead of producing a complete solution, the agent reads the student’s actual work, asks focused questions, explains unfamiliar concepts, and gives one useful hint at a time. It also saves attempts and feedback so students can return to previous explanations and recognize ideas they may need to practice.

Challenges we ran into

One of our biggest challenges was deciding what the website should do and what Superhuman Go should do. We ultimately made Go responsible for the tutoring conversation while our application provides the student’s context and remembers their progress. We also had to make repeated connector calls safe so the same attempt would not be saved multiple times. Another issue we discovered was that normal text validation removed indentation from pasted code, which could completely change the meaning of a Python program. We fixed that so student code, errors, and course excerpts are preserved exactly as they were entered.

Accomplishments that we're proud of

We are proud that SupaTutor supports one consistent learning experience across coding, math, and writing instead of being limited to a single subject. We built the full data flow for assignments, drafts, attempts, feedback, course materials, and progress. We also made the system handle partially correct work, valid alternative solutions, and uncertain situations without pretending the tutor is always right. Our local connection between the MCP tools and the application is working, and all ten automated tests, type checks, and production builds pass.

What we learned

We learned that building a helpful educational agent is about much more than writing a good prompt. The tutor needs the right context, reliable tools, clear limits, and a way to admit when it does not have enough information. We also learned how MCP can connect Superhuman Go to an outside application while allowing Go to remain the place where the student has the tutoring conversation. Most importantly, we learned that educational AI feels much more useful when it responds to the student’s reasoning instead of only focusing on the final answer.

What's next for SupaTutor

Our next step is to deploy the connector securely and complete the full experience inside a live Superhuman Go account. We then want to add user accounts and cloud storage so students can access their work across devices. We would also like to build a VS Code extension and authorized connections for Canvas, GitHub, and lecture transcripts, reducing how much information students need to copy and paste. In the future, a secure code runner could help the tutor verify program behavior while still guiding the student instead of completing the assignment for them.