Matchpiece 不靠人氣配對,奪得黑客松最佳開源 AI 項目挑戰賽第一名
Popularity is a bad way to match people. We built the opposite and won a hackathon.
由五所大學七人組成的團隊開發 Matchpiece,並在 Hacktoberfest 2026 Hack Day Kampala × MUBS 的最佳開源 AI 項目挑戰賽中奪得第一名。平台按用戶的需求和所能提供的幫助進行雙向配對,AI 功能使用 Google DeepMind 的開放權重 Gemma 4 模型。模型無法使用時,平台會改用基於相同配對理念的內置排名;源碼以 MIT 授權開放。
Every hackathon, meetup and conference has the same invisible failure.
Somewhere in the room there is a developer who needs someone who understands customers, and a business student who needs someone who can build. They are ten metres apart. They never speak. They leave with the people they arrived with.
At Hacktoberfest 2026 Hack Day Kampala × MUBS, a team of seven people from five universities decided to build for exactly that failure. We called it Matchpiece: find your missing piece. Out of 8 projects and 67 participants, it won 1st place in the Best Open-Source AI Project challenge, and the team was forwarded to the MUBS EIIC incubator.
This post is about the one idea behind it, because I think it applies far beyond hackathons.
The idea: match by need and offer, not by similarity
Most platforms that connect people rank by one of two things:
- Popularity: followers, likes, connections. The rich get richer, and newcomers stay invisible.
- Similarity: people like you, with your skills and your interests. Pleasant, but useless when what you need is someone different from you.
Matchpiece asks every member two questions in a short chat: what do you need? and what can you offer? Then it ranks people by how well one person's offer covers the other's need, in both directions.
A Flutter developer who needs a designer, and a designer who wants to learn Flutter, are a stronger match than two Flutter developers with identical profiles. That sounds obvious when you say it. Almost nothing online is built that way.
Where the AI helps, and where it is not allowed to
All AI features run on Google DeepMind's open-weight Gemma 4 models, called only from the server. We gave the model a strict rule: use only the facts the member entered. It writes a short headline, bio and skill tags from your own answers, and you accept them or ask for another version. It never invents experience you do not have.
If the model is unavailable, a built-in ranking based on the same need-and-offer idea takes over, so matching never stops working.
From matching to meeting
A match is useless if it stays on a screen. So:
- One tap opens a WhatsApp chat with your match. Numbers are visible only to signed-in members, and anyone can hide theirs.
- Organisers run events in the same app: event page, registration for members and guests, confirmation emails with calendar links, a waitlist that promotes people automatically, and signed QR tickets for check-in at the door.
- Members own their data: export everything as JSON, or delete the profile with all posts and matches. No ad trackers.
Try it, read it, take it
- Live app: matchpiece.tech
- Source code (MIT): github.com/Saifuddin2Ahmed/matchpiece
- Cite it: archived on Zenodo with a DOI, 10.5281/zenodo.23737410
Built by Saifuddin Ahmed, Abubaker Mohamed Adam, Adinan Juuko, Amme Patience Esther, Mupole Uwizeye Alexis, Nabagulanyi Prossy Sherry and Ojambo Emmanuel.
A question for community builders
I keep wondering why so few platforms match people by complementary needs. Is it because it is hard to measure, or because popularity is simply better for engagement numbers?
- If you organise meetups or hackathons, how do you help strangers find the right teammate today?
- Would you trust an AI to suggest who you should talk to, if it only used what you told it?
I'm Saifuddin Ahmed. I led the Matchpiece team, and I'm the founder of Ishraqa Daily, a community on Google Play, the App Store and the Microsoft Store that turns knowledge into impact: developer clubs where people build open-source projects with mentors, languages, reading and listening together, and scholarships and opportunities. I also built Rafu and founded OtwoX Digital Solutions. More of my work is at saifuddin.dev and on LinkedIn.
I drafted this piece with an AI writing assistant. The project, the team and the opinions are ours.
來源:DEV Community · dev.to