AI Competition Radar: August 23-30, 2026

The August 23-30 window produced four result sets with named projects or teams worth a first pass. The new names span a global agent hackathon, a full podium in Oman, a sales-ranked product competition in Hangzhou, and a Chinese national student final.

This issue admits a result when a named winner or placing has appeared on a public page during the week. The competition date and the publication date can differ, so both appear where the record makes them clear.

At a glance

Result set	Named result	What the team built	First diligence question
Global AI Hackathon Series with Qwen Cloud	Five track winners: Quen, Rexgent, Tarmac, Corpus Formation Autopilot, and Lumen. The organizer moved the announcement to August 28, then posted the roster. 1 2	Agent memory, automated drama production, airline disruption resolution, business-formation workflow, and neonatal screening. 3	Which projects have an identifiable team, a usable demo, and a user who can validate the claimed workflow?
Salalah AI Hackathon 2026, Oman	Wafeer first, Flow second, SEER third; Al Yaum Oman published the winner announcement on August 26. 4	A fisheries marketplace, restaurant supply-chain and order management, and a privacy layer for information sent to external AI tools. 4	Which project has a named fisheries, restaurant, or SME partner ready to run a pilot?
Markethon, Hangzhou	HappyPower first, Zizu second, Zai Chang third. The competition ran August 21-23; AI Ping's August 26 account says 59 teams ranked on completed-transaction net revenue. 5	A motion-powered companion device, a subscription AI platform for independent workers, and a virtual study companion. 5	Can each team reproduce sales after the two-day event audience and its distribution channels disappear?
Douyin AI Creator Program 2026 Hackathon League national final, China	Jia Yongshuo, Wang Yunxin, and Ma Yiwei from Hunan University of Technology won first prize in the interactive-space track with Jian Zhi Mei Jia Mo. A local report naming the team was published August 25. 6	An immersive interactive game using AI-generated content and the 2026 World Cup as its setting. 6	Is there a public build, user test, or distribution plan beyond the competition prototype?

Qwen Cloud: five winners, three routes to a live artifact

The Global AI Hackathon Series with Qwen Cloud posted one winner per track after moving its announcement to Friday, August 28. The roster names projects rather than people, so project links are the practical starting point for outreach. 1

MemoryAgent: Quen. Quen treats agent memory as a trust decision: it extracts facts and triples, applies spaced-repetition decay, consolidates unused memories, and checks low-trust memories against live sources before answering. Its submission lists a FastAPI/Python/SQLite backend, Qwen models, a React/TypeScript front end, a live demo, and a GitHub repository. 7

Scout read: the repository and demo support a fast technical diligence call. Ask how the trust score changes user-visible answers, what sources the verification loop accepts, and whether any evaluation used real agent workflows.

AI Showrunner: Rexgent. Rexgent turns a premise or screenplay into voiced, captioned drama episodes. Its LangGraph pipeline covers story development, scriptwriting, casting, storyboarding, model routing, rendering, editing, and export; the project page lists 17 specialized Qwen Cloud models and English/Chinese output. 8

Scout read: request a sample episode with the model-by-model receipt, render time, human-review steps, and rights path for any training or reference materials.

Agent Society: Tarmac. Tarmac has five Qwen agents negotiate airline-disruption seat allocation on a shared ledger. The submission reports a frozen storm scenario in which its full society stranded three protected passengers versus 17 for a greedy planner, and it links a live verification endpoint. 9

Scout read: request the scenario generator and a domain partner's evaluation criteria. A frozen benchmark establishes mechanics; airline operations add live constraints, auditability, and operator control.

Autopilot Agent: Corpus Formation Autopilot. Corpus Formation Autopilot turns a founder request into a cited US business-formation checklist, NAICS selection, and a prefilled filing handoff. The project states that payment locks the filing payload to a hash and a human approves that exact payload. 10

Scout read: ask which jurisdictions and filing paths are live, who owns the legal-review workflow, and how the cited corpus stays current.

EdgeAgent: Lumen. Lumen guides community health workers through a neonatal visit using voice intake, visual examination, respiratory measurement, and a cry recording. It switches between Qwen Cloud and an on-device model; the submission says medication-dose calculations remain in deterministic code. 11

Scout read: treat this as a clinical-validation question. Ask for clinical governance, local regulatory status, data handling, prospective sensitivity and specificity, and the escalation workflow for red or yellow cases.

Salalah: a full local-market podium

Riyada, Oman's Small and Medium Enterprises Development Authority, and Omantel ran the four-day Salalah event with 60 participants across 22 teams. The three winners map directly to regional operating problems: fisheries distribution, restaurant operations, and SME data governance. 4

Wafeer won with a platform that links fishermen directly to customers and markets. Flow placed second with a restaurant and food-and-beverage tool for order handling and supply-chain operations. SEER took third with a tool intended to conceal and protect sensitive organizational information before it reaches external AI tools. 4

The follow-up is local and operational. Wafeer needs evidence of supply onboarding, buyer demand, payment settlement, and cold-chain or logistics integrations. Flow needs a restaurant pilot with before-and-after order or waste data. SEER needs a precise account of its deployment point, policy enforcement, and threat model.

Markethon: transaction revenue as the score

Markethon used completed-transaction net revenue as its leaderboard rather than a judging panel. AI Ping says the event was jointly organized with OPC's Honghub community, ran at Hangzhou City Balcony from August 21 through 23, and put 59 teams through a 48-hour selling period. 5

HappyPower won with a companion device that captures kinetic energy from everyday movement and responds through heartbeat, breathing, snoring, ear, and tail motions. Zizu placed second; its founder describes it as a subscription AI platform for independent workers seeking a self-sustaining income source. Zai Chang placed third with a virtual study companion: a user uploads a parent photo and the application creates a parental avatar that reminds a child to study. 5

AI Ping also gave AmoPaw a zero-cost dark-horse award. The product is a contactless pet cardiac monitor intended to measure respiration, heart rate, and electrocardiogram signals while the pet remains nearby; the team says its orders came through pet-owner communities and on-site demonstrations without advertising spend. 5

Use revenue as a reason for first contact; settle diligence with order count, average order value, refunds, acquisition source, delivery cost, and sales retained after the event. Zai Chang also needs a child-safety and consent review because the product uses a parent's likeness in a supervision workflow.

Douyin national final: a named interactive-space winner

Zhuzhou News reported on August 25 that the Hunan University of Technology team of Jia Yongshuo, Wang Yunxin, and Ma Yiwei won first prize in the Douyin AI Creator Program national final's interactive-space track. The project, Jian Zhi Mei Jia Mo, uses AI-generated content to create an immersive interactive game around the 2026 World Cup. The report says the team refined the game's scenes, interaction design, and technical implementation after qualifying from the Zhuzhou city event and regional competition. 6

The public report establishes the project and placement. A scout should request a playable build, its content-generation workflow, player data, and a distribution plan before treating the award as a company signal.

Result pages still incomplete

Kaiwu: 48-hour AI Maker Hackathon, Fuzhou. A local report says the Fuzhou stop concluded on August 30 with final presentations, a MoneyAI product launch, and investor meetings. Public winner roster: unavailable on the page. 12
Taipei Metro Cup Hackathon. Taipei Metro's pre-event release scheduled the final for August 29 and described prizes for first through third place, a Black Hackers Award, finalist awards, and an Acer AI Maker Award. Public winner roster: unavailable on the page. 13

Scout priority

Quen: open the repository and live demo, then test whether its verify-before-answer loop changes outcomes on a real agent task.
Tarmac: request the benchmark package and an airline or airport operator willing to test the negotiation layer against live rules.
Wafeer and Flow: find the first named supplier, restaurant, or buyer. The Salalah podium has clear local problem statements; public pilot record remains unavailable.
Markethon leaders: separate event-period sales from repeatable acquisition. AmoPaw's zero-ad-spend channel claim and Zai Chang's orders are concrete questions for follow-up.
Lumen: route outreach through a clinician or health-system partner. The product's safety workflow needs clinical evidence before a conventional startup diligence conversation.

The next weekly edition will track newly published winner rosters and project pages from the two incomplete result sets above.