Inspiration

In Homer's Odyssey, a hero's worth is his kleos: what other people hear about his deeds. When Odysseus reaches the Phaeacian court, a bard sings what he has done, and the Phaeacians give him a ship home.

Student engineering clubs have no bard. On our campus, the member who wired the avionics bay every week and the member who came to two meetings both put the same line on their resume: "Member, Maverick Rocketry." Officers track attendance on paper, contribution not at all, and companies looking for local talent can't tell the builders from the bystanders.

We started from tools that help clubs pass down knowledge (projects like Heirloom). We realized the bigger gap is that nobody records who actually does the work. That's Mentor.

What it does

Mentor is a weekly contribution ledger for engineering clubs.

  • Attendance in seconds. Officers tap names, or snap a photo of a paper sign-in sheet. Gemini reads the handwriting and matches it to the roster, and the officer confirms. Each attendance is worth 10 points.
  • Contributions, scored by officers. Each piece of work is rated Low (30), Medium (60), High (90) or Custom (10–100, in steps of 10).
  • Gemini drafts, officers decide. Upload a week's meeting notes and Gemini drafts one entry per member, with a suggested level and the exact quote it came from. Nothing counts until an officer approves it.
  • Live rankings by week, semester and all time, for every club and across the whole college.
  • A company view. Companies browse college → club → ranked members and see each student's full contribution history. They can request to connect (the student accepts) or sponsor the club.
  • Scout. Clubs recruit by posting a link on LinkedIn or Instagram. Gemini scores applicants' resumes against what the club needs, and recorded Mentor work counts for more than resume claims.

A member's weekly score is

$$ W_i = 10\,a_i + \sum_{j} c_{ij}, \qquad c_{ij} \in {10, 20, \dots, 100} $$

where \(a_i\) is sessions attended and \(c_{ij}\) is each officer-scored contribution.

How we built it

  • Frontend: HTML, CSS and JavaScript, with three views (Officer, Member, Company) and Chart.js for weekly points.
  • Backend: Node.js and Express.
  • AI: Gemini API (gemini-3.8-flash) through the @google/genai SDK. Photos and PDFs go in directly, and every call returns JSON against a fixed schema.
  • Database: MongoDB Atlas. Every point lives in one ledger collection, and rankings are computed with aggregation pipelines using $group and $setWindowFields with $rank. $jsonSchema validation rejects any value that isn't a multiple of 10, and a unique index stops anyone from being counted twice in one session.
  • Domain: registered through GoDaddy Registry: [your-domain]

Challenges we ran into

  • Trusting AI with people's records. Our first version let Gemini decide too much. We flipped it: Gemini only drafts, our code checks every roster match, and an officer approves every point.
  • Fairness. Points mean nothing to companies if officers can inflate them. We added an audit log on every edit, blocked officers from scoring themselves, and show within-club rank and percentile next to raw points, because different clubs score differently.
  • Handwriting. Messy sign-in sheets produce guesses. Names Gemini isn't confident about go to a "needs review" list instead of being matched.
  • Scope. We pivoted the core idea mid-hackathon and had to cut hard to keep the demo reliable.

What we learned

  • Structured output (a JSON schema plus our own validation) is what makes an LLM safe to put in front of real data.
  • MongoDB aggregation pipelines can turn a raw ledger into live rankings without storing any totals.
  • The hardest part of a ranking system isn't the math. It's making people trust it.

What's next

  • Percentile-based rankings across colleges
  • Live leaderboards with MongoDB change streams
  • Officer-succession suggestions from each member's record
  • Piloting with engineering clubs at UT Arlington

Built With

Share this project:

Updates

Submission history