Project TYRA is a professional network for Taiwanese researchers and students abroad, and its heart is a mentor–mentee matching program. When I came on as its technology lead, that program was running on a spreadsheet that had stopped scaling. I architected and maintained the platform that replaced it.

Over four cohorts it grew from 274 to 700+ participants a year and made 1,876 researcher connections — enough that the organization now works formally with Fulbright and Taiwanese government education agencies. I built the weekly seminar automation around it (Eventbrite API plus Google Colab) and added POAP-based, on-chain tracking so event attendance could be verified without a sign-up sheet.

The thing I learned building it is that the platform was never the hard part. Matching accuracy was easy to improve; what was scarce was mentor attention. A busy researcher gives you five honest minutes to review the students who picked them, and not a minute more. So the real design problem was to fit the whole review into those five minutes — which is what the dashboard is for.

The mentor dashboard

The matching runs through a Streamlit dashboard I built (MMP_screening) to get mentee information in front of mentors securely and with as little friction as possible. The data lives in Google Sheets; the frontend is a single page.

sequenceDiagram
    Mentee->>Project TYRA: Express interest in mentors
    Project TYRA->>Mentor: Notification email
    Note over Mentor: Profile review (dashboard)
    Mentor-->>Mentee: Brief interview
    Mentor->>Project TYRA: Reply decision
    Note over Project TYRA: Matching algorithm
    Project TYRA->>Mentor: Matching confirmation
    Mentor->>Mentee: Contact for mentoring

It works in two phases:

• Profile review. A mentor requests access and receives a one-time verification code by email. Once verified, they see the mentees who named them — read profiles in a formatted table, download a CSV for offline review, or compare several mentees side by side in a layout borrowed from Apple’s product-comparison pages — then record their preferences through an embedded Google Form.
• Matching confirmation. After the matching algorithm runs, the mentor logs back in to see who was matched with them and confirms next steps through the form.

The hard part was never the matching logic; it was delivering personal data to exactly the right person, in the fewest steps, without leaking it. Two-minute walkthrough →

Built with Python, Streamlit, Google Sheets, Google Colab, and automation APIs, as a volunteer contribution.