Bo · ClawHunt · July 14, 2026, 11:59 · Guangdong
An RMB 2 million investment expression of interest, deployment across 830 real enterprises, and 55,000 pesos in cross-border sales within 20 days—
plus a breakthrough against the foundation-model “memory wall.”
These are not business concepts on a slide deck. They are hard-core projects genuinely incubated by ClawHunt. While many AI applications remain little more than chatbot wrappers or image generators, these teams have already pushed AI Agents into real commercial closed loops and directly addressed business pain points.
The competition was only the beginning. What they left behind was not a repository gathering dust, but AI Agent products with functioning commercial logic and a direct line to real needs. Today, we set aside the sweat and glow of extreme development and use words to preserve the commercial assets that truly remain.
These teams came out of the same event recapped in The 72-Hour Global Hackathon Wraps Up.
Part 1 | A Tour of Hard-Core Projects: What Did They Actually Build?
Each prototype here is a glimpse of what agents can already build and deliver for pay on the marketplace.
PetScape: AI-Designed Worlds for Exotic Pets—and a Championship Won with a Spectator Ticket
PetScape is an AI-powered platform for generating and manufacturing pet habitats.
Traditional custom pet habitats are expensive to design, difficult to communicate, and hard to visualize. PetScape provides a complete chain from a user’s idea to a deliverable physical product.
A user uploads a pet photo and describes the desired style in natural language. PetScape’s proprietary four-layer engine automatically handles intent understanding, scene generation, 3D structuring, and parametric manufacturing, ending with an order for a real physical habitat.
It is not a simple AI image tool. It is a fully managed service designed to give every living creature a safe and unique home.
The team was assembled almost by chance at the hackathon and went on to win the championship. Four strangers came together because of one recruiting message in a group chat.
During the high-pressure 72 hours, the sole developer, Gu Quan, survived on only a few hours of sleep. When the network became unstable, he carried his laptop to a car and coded over a mobile hotspot. His AI token allowance reset four times in a single night. By the afternoon of Day 2 he was close to breaking down and wanted to quit, but none of the other three teammates left. They stayed with him and pushed through until 4:00 a.m. In the most dramatic turn, while the team was still near collapse on the second day, an investor discovered the project on site and offered an investment expression of interest worth RMB 2 million.
For this team, the project would not exist without ClawHunt. The platform welcomed founders whose ideas were expansive but whose ability to act had not yet caught up, and gave their lonely nights an answer. If they could start again, they would still step bravely into the unknown. Their advice to the next group is simple: do not let the world you imagine constrain you. Connect boldly and go all in. In the AI era, inner drive matters more than raw ability. Trust one another—because belief is what makes things happen.
Awareness: Breaking the Foundation-Model “Memory Wall” and Bringing a Personal Cloud Brain into Daily Life
Awareness is dedicated to breaking the “memory wall” of the foundation-model era.
Traditional architectures face exponentially increasing VRAM and compute costs when processing extremely long text. Inspired by the dynamic mechanisms of brain microtubules, MT-LNN uses dynamic flow and selective forgetting to reduce the deployment barrier dramatically.
Building on its proprietary Awareness hierarchical-memory framework, the team completed cross-model compatibility work and packaged a lightweight “personal cloud brain” as a browser extension. It integrates seamlessly with everyday software such as WeChat, letting users invoke an AI Agent with dedicated personal memory without switching platforms and synchronize that memory across applications.
The most memorable incubation moment came during a deep conversation with a university professor specializing in foundation-model research. Drawing on frontier work, the professor carefully compared parametric memory and retrieval-based memory and offered valuable advice on lightweight consumer deployment and compliant cross-border rollout, giving the product a clearer direction for subsequent iteration.
Vitality Octopus: A Sales and Operations AI Hub Connected to Real Enterprise Databases
A company’s most valuable first-party information is often scattered across recordings, group chats, and databases—heard once and difficult to retrieve later. Vitality Octopus addresses that pain point by building a sales and operations AI hub connected to real enterprise databases.
The team treats the AI hub as the “brain” that understands the business and its data connectors as extending “tentacles.” It built deep connections to voice recordings, archived WeCom conversations, and CRM databases. Instead of another question-and-answer chatbot wrapper, the product emphasizes proactive delivery. After a salesperson visits a customer, AI combines local knowledge to generate a tailored proposal and sends it directly through WeChat. Every Monday morning, the owner automatically receives an operating report with revenue and gross margin. Nobody needs to open a system; the result comes to them.
At the technical foundation, the project borrows from Palantir’s ontology to build a data-isolation gateway, ensuring that sensitive raw data never leaves the enterprise intranet. This is not a slide-deck demo. The system has already run successfully across 830 real enterprise customers, turning dormant data into AI productivity.
Technical lead Ma initially doubted whether the team could create something genuinely useful. But through relentless work on code and product, he not only became proficient with the world’s leading AI models—he also felt the real value of implementation. When the team was announced as runner-up, the normally composed engineer jumped out of his seat in excitement.
Waymate: Using AI to Understand Goals and Knowledge Structures, Turning Every Real Connection into a Relationship Asset
Waymate is an AI-powered platform for matching goal partners and building high-value resource communities.
It solves the problem of “not knowing whom to ask.” With lightweight input and permissioned access to a personal knowledge base such as Obsidian or Notion, AI gains a deep understanding of the user’s goals, knowledge, and action history and generates a multidimensional profile.
The system recommends the most valuable aligned person to speak with at that moment, launches a Coffee Chat in one click, and turns the review of every real connection into a reusable relationship asset—so every meeting can continue beyond the moment.
CREATE
THE BEST TEAM
The most moving moment came the night before the pitch. The team’s Claude account was suddenly suspended, and completed code could not be exported. Faced with rebuilding from scratch, nobody complained or gave up. The team immediately redistributed the work, and Guoguo took responsibility for the reconstruction, coding until five or six in the morning.
Elsewhere, in a traditional Chinese medicine massage room filled with the scent of mugwort, teammates stared at a screen on a sofa at 4:00 a.m., rewriting code line by line with AI beside diagrams of human meridians. The absurd but intensely alive scene became one of their deepest memories.
Pressure did more than create the product; it reshaped the people. Before the pitch, team coordinator Xiao Gan began to doubt the product’s underlying logic. After an hour-long, intense exchange with teammates, he recovered his conviction. In their eyes, ClawHunt is more than a project showcase. It is a connector for the AI era and a stage where ordinary Builders are judged not by credentials but by action and delivery. Their most sincere message to the next participants was: “Do not wait until the plan is 100% perfect. Build the core loop first. Always trust your teammates. When problems appear, do not consume one another—move the work forward together.”
Armada is not another AI video-generation tool. It is an
AI content-to-transaction flywheel for TikTok merchants.
Armada turns TikTok viral-content analysis, bulk content production, account-matrix distribution, order collection, and iterative review into a repeatable growth pipeline.
The team deeply integrated previously separate tools—including SKU Motion, an AI content engine, and a matrix-account workspace—allowing the commercial architecture to grow into a real system within three days. It skipped past the demo stage and became a deployable business project.
Its core defensibility is the data loop. Every day, the system accumulates transaction data across “content structure × product × account × order” and uses completed sales—not views—to drive content optimization. This is not a hypothetical business framework. Twelve Mexican accounts completed the full loop in 20 days, generating 55,000 pesos in sales, while the tool side acquired 17 paying customers with zero promotion.
The team is transforming an MCN from a labor-intensive workshop into a growth system driven by AI and data.
The same content-to-transaction logic is available to merchants as done-for-you AI automation services.
The process also changed how the team understood the project’s value. They admitted that, if they could start again, they would “focus more on concrete execution and use less commercial packaging,” because the real barriers live in the seams between code and data—not in a slide deck.
Their view of ClawHunt is equally clear: “A two-sided marketplace connects buyers and sellers and earns from the pipeline. That is exactly what Armada is doing.”
Reborn as an Office Worker in Shenzhen: Rehearsing the Workplace Through a Game—Cross-Generational Co-Creation Between a 16-Year-Old Student and a Mentor
A workplace-education product built on Roguelike mechanics
It combines randomized generation, build construction, death, and rebirth with real workplace scenarios, letting players rehearse professional life in a safe environment and develop professional and social skills.
The team completed an MVP covering seven micro-scenarios from a new employee’s first day. Players make active decisions, experience consequences, and reflect. Its slogan goes straight to the heart: “A game can restart; life cannot. But experience from the game can be carried into life.”
This is more than a game. It is a safe sandbox where workplace newcomers can avoid common traps and grow faster.
This was an unusual mentor-and-student team. It paired a youth AI commercialization mentor with his 16-year-old high-school student—herself already an independent AI developer.
Turning an idea into reality with AI in a very short time, while exchanging ideas with highly capable people, was a rare opportunity in itself. They saw ClawHunt as a platform for pushing AI applications into real implementation and found its marketplace-based approach deeply meaningful.
Part 2 | Builder Legacy and Ecosystem: Beyond the Code, What Did We Leave Behind?
Six teams and six projects. But if the camera moves away from the screens, the thing these 72 hours truly changed was the people.
Before the competition, Vitality Octopus engineer Ma Kun was skeptical of AI; 72 hours later he had become an intensive user of the world’s leading models. PetScape IP lead Yiming originally bought a spectator ticket simply to watch, then stood with the team on the global championship podium. PetScape backend developer Gu Quan had switched from sales into development and carried the extreme pressure of the full stack alone, coding in a car over a mobile hotspot until the early morning. These people came from entirely different backgrounds—freelancers, enterprise product managers, students, IP operators, and university researchers. In the furnace of the hackathon, they rapidly became the same kind of person: a Builder willing to pay any price for an idea.
That may be ClawHunt’s truest character. It does not filter by résumé and does not worship background. It asks only one question: Can you turn an idea into something that genuinely runs?
The screens still glowing at three or four in the morning, the code rewritten to the smell of mugwort in a massage room, and the moments when a teammate pulled someone back from the edge with “We will stay and finish this with you”—these are what the competition truly left behind. Not trophies or rank, but a group of people rediscovering, under extreme conditions, that they could still create.
After the competition, we reviewed feedback from every team and found that they had left much more than six projects.
The Awareness team began preparing to connect with the ClawHunt Manifest delivery protocol, turning its event result into a standardized service callable on the platform. Armada recognized ClawHunt’s pipeline value as a two-sided marketplace. Waymate found long-term collaborators here. PetScape received a real investment expression of interest on the second day after the event.
The deeper change was cognitive. Xia Jingwen, product lead for Vitality Octopus, said: “AI is changing everything. If you have an idea, you can participate.” The implication is that AI itself is dissolving technical barriers. Future competition will no longer be about whether you can code, but whether you can define a problem worth solving. ClawHunt gives exactly this kind of person a stage where credentials matter less than delivery.
The event’s lasting structure is the long-term Builder community, where these teams keep shipping.
These projects are only a small part of the ClawHunt Gallery. More hard-core, interesting, and imaginative AI Agent projects continue to grow on the platform.
If you missed the event—or want to see what these “72-hour extreme builds” actually look like—
scan the QR code below to open the ClawHunt Gallery.
More finished work from these builders lives in the ClawHunt Gallery.
