AI Competition Radar: July 12–19, 2026

Coverage window: July 12 at 8:00 p.m. to July 19 at 8:00 p.m. Pacific time.

Three new result sets are strong enough to act on this week. Gemma for Bharat produced a compact, technically legible podium; Agentic AI Build Week exposed unusually large regional talent density and a real production workflow; and Google for Startups’ AI Agents Challenge surfaced an enterprise-memory project with a named EMEA team. The more familiar mid-July queue is less useful: GitLab Transcend, Mind the Product, and Splunk all passed their scheduled winner dates without a public list in the pages checked for this issue.

Confirmed results to act on

Gemma for Bharat: three buildable India-focused assistants

Event: Gemma for Bharat — GDGoC JIS University 2026, a nine-day, fully online community hackathon organized by GDGoC JIS University. The Kaggle overview records a July 4 start and July 13 close, with 237 entrants, 74 participants, 36 teams, and 37 submissions. 1

Confirmed result: Kaggle’s winners page lists three placements under “Build With Gemma.” Champion Vipra Vedanti | Words of the Wise built a citation-aware HyDE RAG pipeline powered by Gemma 3 for exploring, translating, and querying the Rigveda. First runner MandiMitra built a Gemma 4 crop-market advisor that turns real crop prices into recommendations for Indian farmers. Second runner SwasthyaLink – Vernacular Voice Health Navigator grounds rural health-scheme eligibility and NFHS-5 answers in official data, in the user’s language. 2

Public project links: Words of the Wise, MandiMitra, and SwasthyaLink are linked from the result page. 2

Scout read: This is a better follow-up list than a generic “Gemma app” leaderboard. The three winners point to three distinct diligence paths: retrieval quality and cultural-domain data for Words of the Wise; data freshness, farmer distribution, and trust for MandiMitra; and language coverage plus official-data maintenance for SwasthyaLink. The competition did not publish funding, revenue, or customer traction on the pages reviewed, so the next step is project-level technical and user validation rather than a financing assumption.

Agentic AI Build Week: a large funnel with production-shaped work

Event: Agentic AI Build Week 2026, organized by GenAI Fund in Ho Chi Minh City. Public posts disagree slightly on scale: Su Xian Yue described 2,800 participants and 300 teams, while the ATLAS uGDX announcement described more than 3,000 participants from 64 countries. That discrepancy does not change the signal: this was a large, international build funnel, not a small local meetup. 3 4

Confirmed results:

Retail track: Su Xian Yue’s two-person team won the Guardian Vietnam challenge. Their agent ingests customer feedback from reviews, chats, surveys, and social media across multiple languages; groups material issues; proposes likely causes with supporting evidence; assigns an owner; and tracks before-and-after results. The post does not name the team or give a public demo URL, but it does provide a clear product thesis and a credible workflow target. 3
Built with AWS track: Anshul Roy, a B.Tech student from ATLAS uGDX – School of Technology, was publicly recognized as a winner. The announcement says he competed on an international team through AWS First Cloud Journey; it does not disclose the project name or a demo link. 4

Organizer and ecosystem signal: The Retail-track winner credits GenAI Fund and partner credits from Apify, TinyFish, Langfuse, AWS, and Featherless AI. A contemporaneous report describes the event as a five-day buildathon that concluded in mid-July and highlights projects including Autointro and Aerotrace Tire Intelligence. 3 5

Scout read: The Retail project is the standout outreach lead because it has a measurable operating loop: classify feedback, explain why it matters, route it, and measure whether the intervention worked. That is closer to a deployable customer-operations product than to a demo-only chatbot. The public evidence is not enough to claim production usage, revenue, or incorporation. For Anshul Roy, the useful signal is talent and cross-border team formation; the missing project artifact is the reason to contact before ranking him as a startup lead.

Google for Startups AI Agents Challenge: Synapse makes memory the wedge

Announcement: OXYGY Consulting reported on July 17 that its colleagues Joseph Thomas and Marisha Boyd were named the EMEA Regional Winners of the Google for Startups AI Agents Challenge 2026. OXYGY says the competition attracted more than 1,400 submissions worldwide. 6

Project: Their project, Synapse, is a shared, structured knowledge graph for AI agents. It connects information from meetings, documents, notes, and other knowledge sources so agents can retain and reuse historical context rather than starting each interaction stateless. OXYGY describes the intended shift as moving agents from isolated assistants toward persistent thinking partners. 6

Team signal: Both winners are members of OXYGY’s AI Centre of Excellence, a cross-regional and cross-sector group spanning areas including MLOps and EU AI Act compliance. The public announcement is an employer-backed recognition rather than a funding disclosure; it does not report revenue, customers, or a standalone company. 6

Scout read: Synapse is the most legible enterprise thesis in this week’s confirmed set. The diligence question is not whether “memory” is a useful phrase; it is whether the graph can preserve provenance, permissions, freshness, and useful retrieval across a real organization. OXYGY’s announcement gives scouts two named contacts and a concrete wedge, but no public demo or repository was linked.

Results that need a follow-up, not a ranking

Mid-July had a dense cluster of scheduled winner dates. The pages below establish that the dates passed; they do not establish a winner list. They remain useful as monitoring targets, but they should not be mixed with the confirmed podiums above.

GitLab Transcend

The official Devpost schedule set winner publication for July 10 at 5:37 p.m. Eastern time after judging ended July 8. The dates page checked for this issue contains the schedule but no public winner names or project list. Until GitLab publishes a result page or a verifiable announcement, do not treat a project-gallery favorite as a ranked winner. 7

Mind the Product: Everyone Ships Now

The official schedule set winner publication for July 15 at 9:00 a.m. British Summer Time. The event page describes the World Product Day build challenge, but the schedule page checked here does not expose a public podium. The gap matters because the event’s 1,300-plus participant and 300-plus submission signals are useful only as funnel size, not as evidence that any particular project won. 8

Splunk Agentic Ops

Splunk’s official schedule set winners for July 17 at 2:00 p.m. Pacific time after judging closed July 10. The public dates page checked here does not include winner names. Keep this on the watchlist for agent observability and operations tooling, but do not promote a submission to “winner” without the missing announcement. 9

Google Cloud Rapid Agent

The previously tracked public Devpost endpoint for Google Cloud Rapid Agent now returns a “page does not exist” response in this run. Because the result page could not be verified, this issue does not quote a winner, participant count, prize figure, or project list for the event. That is a source-access gap, not evidence that the event had no winners. 10

Next scout calendar

July 20 — Global AI Hackathon Series with Qwen Cloud: submissions close at 2:00 p.m. Pacific time. The official Devpost page advertises $70,000 across five tracks; judging runs July 28–August 11 and winners are scheduled for August 17. This is the immediate action item for scouts who want to see fresh agent projects before the queue closes. 11
July 24 — FIND EVIL!: winners are scheduled for 12:00 p.m. Eastern time after judging through July 17. The event is a security-focused AI competition, so the result is a useful complement to general agent-builder coverage. 12
July 31 — H0: Hack the Zero Stack: winners are scheduled for 2:00 p.m. Pacific time after judging ends July 24. Keep the event in the calendar rather than treating the June submission close as a result date. 13
August 4 — UiPath AgentHack: the official schedule sets winner publication for 6:00 p.m. Eastern time. The event remains a future result, not a missed July announcement. 14

Outreach order

Start with Synapse if the mandate includes enterprise knowledge, agent memory, or governance. The EMEA regional win, named OXYGY team, and specific knowledge-graph thesis make the first diligence step unusually clear, even though a public demo is missing. 6
Follow the Agentic AI Build Week Retail winner for customer-operations infrastructure. Ask for the demo, deployment environment, evaluation set, and evidence that the before-and-after loop is more than a workflow mockup. 3
Use the Gemma for Bharat write-ups as a technical screen. Words of the Wise, MandiMitra, and SwasthyaLink each expose a different combination of retrieval, structured data, and vernacular access. Their Kaggle write-ups are available now; the next diligence question is usage beyond the hackathon. 2
Do not close the mid-July queue until the missing result pages are fixed. GitLab Transcend, Mind the Product, Splunk Agentic Ops, and Google Cloud Rapid Agent are precisely the kind of events where an inaccessible winner post can create false negatives. Track them separately from confirmed results and update only when an organizer or verifiable team announcement closes the gap. 7 8 9