After reviewing the 59 passionate teams from the Xiaohongshu Hackathon, we would like to recommend these 6 | This week's AI project recommendations

This issue's AI project recommendations all come from the Xiaohongshu Hackathon. This week, I served as a judge at the largest hackathon in Xiaohongshu's history. “Largest in history” is a tagline I added; this is actually Xiaohongshu's first 48-hour hackathon focused on AI. Last year, they held an independent developer competition, but the difference is that one featured more mature products competing, while the other focused more on creativity, team formation, collaboration, and the ability to hack together within 48 hours—this is all about “evaluating people.” This edition was indeed large in scale, with a spacious venue. Some participants who have attended numerous hackathons stated that this was the largest hackathon they had ever participated in. However, many participants still complained about the on-site internet, which seems to be a problem that all hackathons have yet to solve. This also adds more challenges to the 48-hour development period. 48 hours, temporary team formation. The competition was divided into hardware and software groups, with judges from different fields watching demos, listening to pitches, and interacting with teams on the final day. The top 10 selected through scoring could showcase themselves on the final stage, ultimately determining the top three in each category and an overall grand prize. In the process of interacting with these teams and scoring their products, one can discover the unique aspects of Xiaohongshu's hackathon. “Unlike many hackathons, I found that at the beginning of the exchanges, no one was holding back; they all wanted to share their project ideas and thoughts with you,” said one participant during the discussions. Xiaohongshu put considerable effort into selecting and inviting different participants to compete, most of whom are “building in public” on the Xiaohongshu platform. They are users of today's various latest AI tools and application products, and many of them are founders of startups themselves. They are generally very active on Xiaohongshu, promoting their products and engaging directly with users. This means that many ideas from the participant group are representative samples of today's AI application entrepreneurial directions and overall user needs. After a day of in-depth discussions with the generally post-00s and even post-10s participants, here are my observations and a few projects I want to recommend These projects are not all those that ultimately receive official awards, and they may not necessarily become standalone products. However, most of these projects come from teams composed of multiple entrepreneurs, which means they have greater reference value for today's AI application directions and a higher likelihood of continuing or becoming part of a more mature product. "Data" Everyone knows that data will be the key moving forward. Many of the projects I have seen essentially aim to gain more user data permissions through various creative means. For example, cleverly using the "hanging beam" method, through camera + visual model capabilities to identify your "bad behaviors," and then using methods such as electric shocks or AI loudly "abusing" you to help you quit bad habits with an AI headband. The gimmick attracts users, but the first-person perspective data is the real treasure. Another example is designing a jellyfish-shaped bedside lamp, combined with millimeter-wave radar, to collect more data such as your body posture, along with an agent product, to provide you with more complex health advice based on sleep. "The sleep economy's agent." Project Recommendation: Wuwei Creation This team aims to address the data island problem during the AI hardware explosion process by creating an open agent framework to process all data from your Plaud, Looki, Ray-Ban Meta, etc., in one place, allowing these different hardware devices to access each other's data for better interaction. They have also developed a hardware solution that allows users to DIY their own AI hardware in a modular way, using this open agent framework to connect all hardware data. Among them, building a unified agent framework for different hardware is an interesting idea. Moreover, it does not unify all data into language data for processing by language models, but aims for end-to-end processing. Physiological "signal" data such as heart rate and blood pressure have not yet been truly "integrated" and processed in current AI applications. They want to solve these problems through a unified agent framework, incorporating these raw, unstructured signal data into the scope of end-to-end processing to achieve deeper information mining and more real-time interaction. Imagine a scenario where the agent knows through the phone that the user has an early flight the next day. In the early morning, if the bedroom camera detects that the user has not yet gotten up, it will call an external sound control to wake the user up. In terms of business model, they believe that the hardware itself will not be the main profit point, and they will mainly rely on subscription-based software services for revenue. Team members include Wu Xinquan, He Yufan, Zhao Qiutong, and Li Yue, among whom are several founders with backgrounds in large companies who are now starting their own ventures. Urgently Need to Break Free from the Imagination Limitations of Lobsters The competition is divided into two tracks: software and hardware. My feeling is that the ideas in hardware are overall more appealing than the products from the software group. One important reason is that software is still trapped in the trend of "lobsterization," where the youngest and most creative individuals today are still too easily limited in their imagination by "lobsters." Various different descriptions of "shrimp" have already lost much of their interest. On the contrary, some projects originating from hardware demonstrate greater imagination. Project Recommendation: Grounded A project called Grounded aims to create a one-stop hardware design platform where users can describe their desired personalized hardware products in natural language, and the Agent will automatically complete the process of recommending solutions, circuit design, ordering, and more. For example, if you are a child who wants a smart watch that looks different from your classmates' genius watches, you can draw or describe the appearance you want, and this agent will handle all subsequent processes, ultimately providing you with easily assembled components to create your own unique genius watch. The team consists of He Yongxian and Xia Liwei, one of whom is the CTO of a startup that has already secured early financing, while the other has extensive experience in the entire hardware design process. This is an excellent combination. It is understood that this project will continue and plans to launch a formal version. Interaction Still Has Great Potential In this context, several software products that caught my attention all start from interaction, including the concrete implementation of generative UI brought by AI capabilities, and how to use AI to handle familiar human elements with more familiar interactions. Two corresponding projects worth noting are Attune, which creates "Attention UI," and Monoslides, which aims to create better-looking PPTs through "serving" AI. Project Recommendation: Attune It aims to reconstruct the web browsing experience by tracking users' dynamic attention. Currently, it exists as a browser plugin that analyzes and gathers interactive elements on the page around the mouse, providing additional features based on user interests (such as data visualization). When users are browsing, it automatically hides distracting information. You can understand it as a more atomic version of an AI hoverball; when you are browsing certain web pages and see information you want to learn more about, this plugin will pop up to allow you to complete the action you want with one click. "Don't go looking for various UI interfaces; let the UI interface find you." The team members are quite interesting: Edward Luo Guorong, with a design background, previously developed the popular dynamic island AI "Vibe Island," which allows you to interact with all your AI Agents on the dynamic island of Mac; Involved in AI application entrepreneurship, Mo Shaozheng and Wu Haiwei, who previously won awards at the Xiaohongshu Independent Developer Competition, have multiple products that already have a considerable number of loyal users. This product also attracted Cao Xi from Monolith to inquire on-site: What if teams like yours, coming from large companies and developing products independently, do not need financing? Project Recommendation: Monoslides At first glance, this seems like "yet another AI PPT," but as a hackathon project, its main exploration is that the process of creating content like PPT should also be CLI-based: Currently, large models lack the ability to read XML structures by decompressing PPT files, leading to a lack of understanding of quality design. They aim to improve generation quality through a three-layer infrastructure—using SVG, familiar to AI, as an intermediary format, deconstructing design with "Design Tokens" to activate large model cognition, and developing a Linux sandbox environment within the browser to reduce Agent costs. The team includes Xue Lai and Zhang Jixiang. Xue Lai is already a serial entrepreneur. Project Recommendation: Vibe Center A hardware aiming to create a new interactive entry point in the AI era, simply put, it aims to eliminate the keyboard and create a new hardware input device centered around AI usage. The prototype created in 48 hours looks like this: Screen, turn knob, several buttons, a microphone. It integrates quick commands, voice intent input, multi-AI process monitoring, and a local knowledge base (memory). Among them, unifying the memory of all AI products onto this edge device is also an interesting direction. This product is also the product direction of team member Wang Binghan's newly established startup. He previously worked at Anker, and the next plan is to make this product look better. He stated that the new product will be ready soon. Team members also include Li Zihao and Li Xingtin. In other words, the potential for AI to change interactions has not yet been fully realized, and these are much more interesting than making various shrimp. be Founder, not just builder This hackathon had many participants who are already founders, and it is also bound to see some participants who may ultimately decide to start a company because of this hackathon. A team called Mira impressed me. Project Recommendation: Mira This is a team composed of three groups of people, who completed everything from idea to hardware to final presentation from scratch in 48 hours. The team includes Wang Jianle, Wang Ying, Hu Litong, Cui Yao, Wang Junkai, and He Yuan. They started from scratch to create a Pixar-style jumping light called Mira. You can interact with it, and you can connect more data to make it understand you better. For example, one scenario they can achieve: its corresponding agent can obtain your email permissions, it knows you have been waiting for an offer, and when you receive and open this offer, Mira will understand this feeling, start dancing happily and flashing lights, celebrating with you. These individuals stayed awake for 24 hours "extreme modeling, six iterations, using two Tofu A1 printers to continuously work for 48 hours to complete the appearance. Then for the hardware part, they changed six sets of hardware solutions, burned out servos, damaged batteries, and finally completed Mira just before the 48-hour deadline." They filmed a demonstration video, but the lamp fell off the table while twisting, breaking its joints, and ultimately could not be demonstrated in real-time. They were one of the top 10 teams for the final presentation, but ultimately did not win the grand prize. In fact, this team has already shown a structure resembling a small startup, completing an unprecedented task in 48 hours in the form of an AI-driven organization. This is also a typical team that emerges naturally after Xiaohongshu changed to a "people-focused" logic. The future is promising. Narrative and Xiaohongshu's Incubator Attributes In addition to these exciting projects, throughout the hackathon, one can feel the potential of the Xiaohongshu platform in the AI wave. "What if I post my product on Xiaohongshu?" This has become the default way of thinking for every team, meaning that these developers, who are already or potentially will become founders, start considering the issue of growth through narrative and dissemination at the very first moment of designing a product. At last year's independent developer conference, many relatively mature products were updated based on user feedback obtained through user operations on Xiaohongshu. This year, many hackathon participants further used Xiaohongshu as a place for rapid idea validation and trial and error, thus determining product forms. It has become the first link that these founders cannot avoid when making judgments on key entrepreneurial issues. Xiaohongshu has already become an important distribution channel for AI startups and products, as well as the default place for product trial and user operations (possibly without exception). This is a positive change. A new wave of entrepreneurs generally realizes that storytelling ability is a very important skill. You can see that every product tries to establish its own narrative. However, after watching the demo presentations of the top 10, it is still evident that compared to the exchanges we participated in Silicon Valley and some we hosted ourselves, these young people, who are often founders of startups, still have some gaps in how to tell their products to a larger public and "sell" their dreams. The pocket guitar product that ultimately won the grand prize, besides having a great hardware design, was also the team that presented most like a real product launch event, which greatly helped in gaining recognition from investors and judges. From the perspective of the current schedule setup and scoring criteria, Xiaohongshu is clearly looking for potential investment targets. It evidently considers narrative ability as an important indicator for evaluation. It can also provide value that other platforms do not have through community-driven traffic support and operational method sharing, which all contribute to Xiaohongshu developing incubator-like attributes. Therefore, if one day Xiaohongshu's hackathon evolves into a YC demo day or a16z speedrun event, it would probably not come as a surprise to anyone.