A product manager and full-stack developer spent six months using AI to build and test a campus-focused social mini-program. With prior access to a community of more than 300,000 users, existing traffic and social-product experience, the author independently developed its architecture, core infrastructure and features including AI search, AI recommendations, an AI “wingman,” relationship tests, payments and referral mechanisms. The project showed that one capable person can use AI to produce and operate a product that previously might have required a team of five or six people. The initial concept was a “search engine for people”: users would describe the person they wanted to meet, while AI searched profiles and helped matched users start conversations. In practice, about 80% of searches focused on a small set of superficial attributes, and the system mainly served the searcher’s wishes without solving whether the other person wanted to connect. The AI wingman could improve conversations after mutual interest existed, but it could not create interest before a conversation began. The team then returned to a conventional mutual-selection model, retaining AI-assisted recommendations and conversational interaction. Early results looked promising: registrations quickly exceeded 100 after promotion, next-day retention surpassed 35%, daily active users approached 1,000 at a peak, and some users paid and returned repeatedly. Longer-term analysis showed that only about 5% of the promoted audience would register and place themselves in the selection pool. The initial growth wave likely converted high-intent users from an existing campus pool rather than creating a self-sustaining growth loop. The team estimated that the pool would need to grow by 10 to 20 times, potentially requiring tens or hundreds of times more resources and eventually around 100,000 real registrations and at least 10,000 daily active users. Acquiring and maintaining trustworthy, balanced and active users would create costs comparable to a larger social business, while the product lacked a strong differentiation from established dating and campus-social services. The mini-program was therefore paused because usage and revenue did not demonstrate a sustainable return on investment. The author concludes that AI lowers the cost of building and validating ideas, but not the difficulty of solving demand, trust, user acquisition and monetization. Future experiments may focus on AI-assisted, low-pressure offline group formation around activities, although that direction would require greater local density, safety controls, fulfillment and operational resources.
