After leaving Moonshot AI, he leverages AI technology to help people find their ideal romantic partners and has received investment from Xu Xin | New Emerging Project

Text by Wen Lihong

Edited by Zhang Yuxin

On the eve of the launch of Kimi K3, Zeng Xunxun did the math: the value of the stock options he gave up when he resigned had risen 10 times during his time away from the company.

Before leaving Moonlight, Zeng Xunxun was the technical lead for Kimi's AI search. In August 2025, he left to found "Perfect Match Technology", with its core product "Perfect Match" — an AI-powered dating and relationship matching application. Its core logic is to integrate AI search and matching capabilities into the process of "finding people", using AI to improve efficiency in three key stages of the dating scenario: profile collection, intelligent matching, and communication assistance.

Shortly after the company was founded, Zeng Xunxun secured a $2 million angel investment from Xu Xin of Capital Today. He later learned that before meeting him, Xu Xin had already screened multiple rounds of projects in the "AI for people search" space. At the time, she was focusing on two directions: "AI job hunting" and "AI partner finding".

In the recruitment sector, she spoke to almost every notable domestic startup team in the field, but did not invest in any of them. Xu Xin's reasoning was that none of these teams had a deeper understanding of recruitment than the previous generation of product managers. No team could provide her with an unexpected answer about which links AI could create incremental value in, or how it could restructure matching efficiency.

For the dating sector, Xu Xin only met with the Perfect Match team. Before investing, she mentioned that she was looking for a founding team whose understanding of "AI search and matching for people" surpassed her own, and that of any company she had previously invested in. If it did not exceed expectations and only matched existing standards, there would be no value in placing a bet.

Currently, the Perfect Match mini-program is online, and the full App is about to launch.

Zeng Xunxun, Founder of Perfect Match, speaking at the 2026 World Artificial Intelligence Conference (WAIC). Source: WAIC Official

A Startup Built on Familiar Technical Foundations

Zeng Xunxun believes he is entering a market with proven real demand. He cites data from several similar products as reference: the first platform charges 1288 yuan for an annual premium membership, while the second sees an average lifetime payment of about 800 yuan per paying user.

An annual fee of 1288 yuan falls in the highest pricing tier for consumer subscription products. Compared to the 200-300 yuan annual memberships for video platforms and 100+ yuan for music platforms, the fact that users are willing to pay 1288 yuan for a dating service shows that "finding a partner" ranks extremely high in users' value priorities.

At the same time, a common longstanding concern among users of dating platforms is that after paying high fees, they still struggle to accurately and efficiently find the right person. On this point, Zeng Xunxun's judgment is: The technical capabilities of the previous generation of dating products cannot deliver the "precise matching" service. These platforms built a huge user pool to gather people with relationship-seeking needs, but when it came to the actual process of finding a partner, users still had to invest massive amounts of time and energy. Users had to browse profiles, analyze information, communicate, match, and filter candidates over and over, while recommendation systems did almost nothing to help them more accurately identify compatible people.

The core logic of "Perfect Match" is precisely solving this "precise matching" problem, which is highly aligned with Zeng Xunxun's previous work at Kimi.

While serving as the technical lead for AI search at Kimi, Zeng Xunxun spent every day leading his team to research how AI could help users accurately match information, and explore the commercial value that could be created from this simple core action.

For a long time, he observed a pattern in AI search: in the general search field, as technology improves, search results get better and answer quality rises significantly — but at the same time, the computing power costs also become higher. However, it is difficult to get users to pay for a single search. "Right now, all AI search services are offered for free, users can use them without paying token fees, because everyone thinks users shouldn't have to spend money on this, and no one is willing to pay," Zeng Xunxun says.

But for AI search products dedicated to specific verticals (such as healthcare, finance, law, etc.), the situation is very different. These search services often integrate a large number of private databases to help users search and match, and users are willing to pay for these targeted results.

Zeng Xunxun believes this is not only due to differences in technical performance and costs, but also because the answers in these verticals are much more closely tied to people's vital interests: a professional medical answer could save a life, and a professional legal consultation result can directly impact major decisions.

The high premium in these scenarios clearly shows that it is not that users are unwilling to pay for "information" — it is that the value of information in general scenarios is too low. The same AI technology cannot find a profitable path in general search, but when applied to specific search scenarios like law and healthcare, it can find opportunities for commercial monetization.

Following this line of reasoning, he concluded that "finding people" is another high-value scenario for search. Two typical use cases stand out: recruitment, and dating and relationships.

In Zeng Xunxun's view, for AI, finding people is essentially the same as high-precision information matching. The technical actions required for AI search and dating matching are completely identical: "Understand intent — analyze and match requirements — output search results". Kimi works by taking a user's question and matching relevant information; Perfect Match works by taking a person's personal criteria and expectations for a partner, then matching them to another person. The target of the search has changed, but the underlying logic remains the same.

Using AI search to power dating matching, he breaks the process down into three stages, each corresponding to a specific bottleneck in traditional dating platforms.

The first stage is profile collection. On dating platforms, writing a self-introduction is a standard task. But not everyone can describe themselves clearly, logically, and comprehensively, so a large number of user profiles on these platforms are nearly blank, leaving the system with no effective information to generate recommendations.

In Zeng Xunxun's view, this is not a problem of user willingness — it is a problem of user capability. On "Perfect Match", the design team built an "AI Matchmaker" to help users complete profile collection. After registering, users have a 20-minute voice call with the AI Matchmaker, which asks personalized questions. Users do not need to carefully organize their thoughts, they just need to answer truthfully. After the call, the AI automatically generates a high-quality self-introduction, sorts out the user's relationship expectations, values, and other basic information required for dating matching, which is used for subsequent demand analysis and matching.

Interface demonstration of the "AI Matchmaker" feature on Perfect Match, photo provided by the interviewee

The introduction of AI technology lowers the barrier for users to create comprehensive, high-quality profiles, removing the requirement to "be able to express themselves precisely". After the Perfect Match mini-program launched, Zeng Xunxun and his colleagues found that the average length of user profiles was 463 words — while their previous manual survey of leading dating products found the average user profile was only 132 words long.

The second stage is the precise matching algorithm. Previous generation products relied on structured tags like age, height, and education to generate recommendations. But in reality, the core factors that determine whether two people are compatible are tied to their values, life experiences, personalities, and lifestyles.

All this information exists in the text users write about themselves, but traditional recommendation systems could not analyze or utilize it. Even after being matched based on tag information, users still had to sift through countless people with mismatched marriage expectations from a huge pool, and finding the right person remained a draining, difficult process.

Zeng Xunxun clearly recognizes this: previous generation recommendation algorithms could not process natural language — until the advent of large language models. Perfect Match is currently training a dedicated matching model, which takes as input the user's full profile, including the public self-description and the raw transcript of their conversation with the AI Matchmaker. After the model understands the content of two users' profiles, it analyzes their deep-level personal information to enable far more accurate subsequent matching.

The third stage is communication facilitation. In dating scenarios, there are some critical questions that need to be clarified before a relationship becomes serious — such as family assets, debts, relationship history, health status, and expectations around marriage-related expenses. In the early stages of getting to know someone, these questions are impossible to bring up directly. If you wait several months into the relationship to ask them, the sunk cost will be so high that some people will be unwilling to face any potential disagreements.

To address this, Perfect Match has developed two features: AI Doppelganger and AI Strategist. The former allows users to ask and answer questions anonymously — the recipient does not know exactly who asked the question, they only know that a potential match wants to learn about a specific detail of their life, eliminating the awkwardness of one-on-one information checks. The latter, the AI Strategist, is responsible for providing conversation topic suggestions and guidance on appropriate attitudes during real-person interactions.

Choosing Between User Scale and Community Culture

Zeng Xunxun has observed that many dating platforms have to constantly waver between "user growth metrics" and "maintaining a healthy community culture".

In order to expand their user base, platforms are often reluctant to position themselves strictly as a "serious relationship-focused platform", and instead generalize the types of connections they facilitate, using slogans like "help you meet new friends" or "help you find a date". This blurs the boundaries, allowing users with all kinds of different intentions to join. As a result, users with conflicting goals are mixed together on these platforms, and in the end, no one is satisfied.

Zeng Xunxun shared a thought with "Intelligent Emergence" that sums up his fundamental judgment on this issue: "If you believe you are building a serious dating app, you should uphold that atmosphere in every single aspect of the product. None of the past products did this well — they all tried to pursue multiple goals at once, and ended up achieving none of them properly."

His choice was to abandon that "trying to have it all" mindset, and clearly define Perfect Match as a serious, marriage-oriented dating platform. Centered on this positioning, he designed filtering mechanisms across all stages of the user journey to let non-target users opt out during the registration process.

Perfect Match has implemented multi-layered filtering mechanisms: real-name authentication, facial recognition, a roughly 20-minute voice call with the AI Matchmaker, and an AI profile evaluation (users can only go online if their profile scores above 70). Users must pass all these steps in sequence to access the platform. The platform also verifies the marital status of registered users, and directly blocks anyone who is already married. In the first week after the product launched, more than 50 married users were prevented from registering.

In addition, Perfect Match has designed a strict "1-on-1 matching mechanism". After a user enters a matching session, they can only chat with one matched person at a time. They can only unmatch and start getting to know someone else after confirming the current match is not suitable. "This is the most important product tradeoff we made to maintain a serious community atmosphere," Zeng Xunxun says. "This mechanism eliminates the suspicion and game-playing that is common in casual social platforms, while ensuring that every match is treated seriously by users."

All these rules are built on a core logic that Zeng Xunxun firmly believes in: "If someone is only looking for a short-term relationship, they will not be willing to reveal so much real personal information on a platform, nor will they go through so many filtering steps."

The community atmosphere in turn shapes user behavior. Zeng Xunxun has verified this insight: a good community culture is not enforced by rules alone — it is cultivated through intentional filtering mechanisms. When the entry threshold is high enough, users who meet the standards will naturally form a positive atmosphere after joining the community; and that atmosphere will in turn guide the behavior of new users entering the platform, creating a virtuous cycle.

In Zeng Xunxun's view, AI can improve the efficiency of precise matching by accurately analyzing users' relationship-seeking needs. But whether a serious dating community can succeed ultimately depends on how strict the filtering mechanisms are, and whether the business model aligns with the product's core goals. These are things that AI cannot solve — they require human insight and deliberate implementation. A team that only understands technology will find it difficult to pull this off.

"Marriage Guarantee": A New Attempt at Outcome-Driven Service

In Zeng Xunxun's view, the fundamental reason why previous generation dating products failed was not entirely backward technology, but a fundamental conflict between their business model and product goals. In the past, almost all dating products operated on a subscription model, charging membership fees on a monthly or annual basis. This meant that the longer a user stayed on the platform, the more revenue the platform made. As a result, platforms had no real incentive to help users find a partner as quickly as possible.

"In this scenario, platforms do not truly want users to succeed in finding a partner. They design all kinds of ways to get users to pay, and after squeezing out every last bit of value from them, users end up disappointed and leave," Zeng Xunxun says.

So he redesigned the service delivery logic for the dating platform. When users pay, they are not buying a fixed period of access to platform services — they are paying for a clear "marriage outcome" as the core service deliverable. Specifically, he has planned a "Marriage Guarantee Membership" program, with a core rule: After users pay a 2000-yuan membership fee, if they do not obtain a marriage certificate within 3 years of joining Perfect Match, the platform will refund the full payment.

This model went through three iterations of refinement.

The first version was "guaranteeing that users will meet at least one person". Users could send messages to each other, chat for a few rounds, and then arrange to meet in person — as long as the in-person meeting happened, the service was considered delivered. If a user did not meet a single person, they would not be charged at all. He discussed this plan with investors, and the feedback was that users might cause disputes: maybe two people exchanged WeChat contacts but never actually met, or they met but had a very unpleasant experience, leading the user to refuse to acknowledge the service was completed — it was too risky.

The second version was "guaranteeing that users will enter a romantic relationship". This had even bigger problems, because being in a relationship is impossible to verify — anyone who did date could claim they never did, and the rate of users refusing to pay would be close to 100%.

Eventually he realized that marital status is the only verifiable, non-falsifiable service outcome. So he defined the service deliverable for paying users as "delivery based on a successful marriage outcome".

The "3-year" time limit was not chosen arbitrarily. His team distributed around 150 questionnaires to people queuing up to get marriage certificates at civil affairs bureaus in Nanshan and Futian districts, Shenzhen, hoping to learn how these newlyweds met, and how long they knew each other before getting married.

"Excluding natural relationships like classmates, colleagues, and childhood sweethearts, 70% of people who met through introductions or online platforms after starting their careers got married within three years," Zeng Xunxun explains. "If they were a good match, they got married; if not, they split up. Three years is a very natural timeline."

Regarding concerns that users would "maliciously delay getting married for three years just to get a refund", his judgment is: "For Chinese people, marriage is one of the most important events in life, and the expenses involved are far higher than 2000 yuan. Compared to the significance of getting married, 2000 yuan is not enough to influence people's decisions — even choosing an auspicious wedding date costs more than that."

Currently, Perfect Match's product is still in its early stages. Whether the AI Matchmaker can consistently generate high-quality profiles, whether the dedicated matching model can reach 80% accuracy, and how many users will get married after three years — all of these things still need time to verify.