Cactus x Google DeepMind Hackathon
About This Hackathon
This is a multi-city, one-day global hackathon organized by Google DeepMind, Cactus Compute, AI Tinkerers and AI Nexus.
Across different cities, builders will simultaneously explore a new frontier of AI development:
Agentic systems that run locally with on-device using FunctionGemma on Cactus Engine, with seamless fallback to Gemini in the cloud.
This isn’t a slide-deck hackathon. It’s a room full of engineers shipping real systems — mobile, desktop, and hybrid — that push inference closer to the user without giving up access to frontier models when needed.
If you care about:
Local-first AI
Latency, privacy, and offline capability
Agentic workflows that actually ship
Bridging edge + cloud in production
…this is your room.
About Cactus Compute
Cactus Compute is an engine for running AI locally on-device, enabling developers to ship fast, private, and reliable AI experiences without relying exclusively on the cloud. Cactus makes it possible to run complex agentic workflows on mobile devices and Macs, with seamless escalation to frontier models like Gemini or Claude when needed.
At this hackathon, Cactus empowers builders to rethink where AI runs — unlocking local-first, hybrid AI architectures that reduce latency, improve privacy, and work even when connectivity is limited. If you’re building agentic products that need to feel instant and dependable, Cactus is the runtime layer that makes it possible.
About Google DeepMind
Google DeepMind is a world-leading AI research lab focused on building general-purpose, safe, and scalable AI systems. From frontier models like Gemini to breakthroughs in multimodal reasoning, planning, and tool use, DeepMind’s work powers some of the most advanced AI capabilities in production today.
At this hackathon, DeepMind is supporting builders who want to move beyond demos — exploring how Gemini can be integrated into real agentic systems, especially when paired with local execution and intelligent cloud fallback. This is about shipping systems that reason, act, and adapt in the real world.
What You’ll Be Building
Builders will create agentic applications that combine:
FunctionGemma for fast, local execution on mobile devices and Macs
Cactus Compute as the runtime for running AI locally on-device
Gemini APIs as a cloud fallback for more complex reasoning and heavy tasks
The goal:
Design systems that intelligently decide where computation should happen — locally or in the cloud — based on context, cost, latency, and capability.
Example directions (not constraints):
Offline-first AI assistants with cloud escalation
Local agents coordinating tools and workflows
Privacy-preserving AI apps that minimize data exfiltration
Hybrid inference pipelines across edge + cloud
New UX patterns unlocked by on-device intelligence
You’ll ship something real, demo it live, and get feedback from peers who also ship.
Who This Is For
This hackathon is curated. Attendees are actively building AI systems in production or near-production environments.
You’ll be working alongside:
Senior software engineers
Technical founders
ML / infra engineers
Researchers translating ideas into products
Expect hands-on debugging, architecture debates, and real trade-off discussions — not surface-level networking.
Schedule
8:30 AM — Doors open, breakfast & networking
9:00 AM — Welcome + technical overview from Google DeepMind & Cactus Compute
10:00 AM — Hacking begins
10:30 AM — Team formation deadline
5:30 PM — Project submissions due
6:00 PM — Preliminary judging
7:00 PM — Final demos
8:00 PM — Winners announced
8:30 PM — Event concludes
How Teams Work
Teams of 1–4 builders
Solo builders are welcome!
You can arrive with an idea or form one on the spot
All teams must submit:
A working demo
A short explanation of architecture & trade-offs
Clear use of FunctionGemma and/or Gemini
What You’ll Get
Direct access to DeepMind and Cactus Compute technical context
A high-signal room of builders who ship
Real-time feedback on your architecture and design decisions
Exposure for standout projects across the AI Tinkerers network
Prizes and recognition for top teams (details announced on-site)
Most importantly:
You’ll leave having built something real — and learned patterns you can reuse immediately.
Register
Space is limited to 150 attendees per city, and registration is reviewed to maintain a high-signal room.