Garrick - Google DeepMind x Cactus Compute Global Hackathon
AI Tinkerers - San Francisco
Hackathon Showcase

Garrick

Team led by an Algoverse AI researcher with 15+ years of experience in C++20, PyTorch, low-latency multimodal pipelines, and quantitative finance.

1 member Watch Demo

My agentic workflow, SwissblAIz, is a deterministic tool calling decision engine kernal that leverages Cactus Compute Runtime on-device for FunctionGemma 270M-3.1B LFM models with Gemini Flashback & Pro could fallback reasoning on complex or hard queries. It also used Python, FastAPI, Tyescript ,Web Assembly, Cactus SDK, local vector store, custom routing controller, structured logging, and confidence based evaluators. Together these tools enable dynamic escalation, agentic utility, and local-first performance by combining deterministic on-device execution with intelligent cloud tool call fallback. The result is a state of the art hybrid agent that is fast, private, resilient and deeply aligned with the next generation of edge-cloud hybrid agentic AI systems.

None

AI Tinkerers Cactus Compute Cactus Compute Runtime (on‑device inference engine for FunctionGemma Cactus SDK FastAPI FunctionGemma 270M (local function‑calling model) Gemini Flash / Gemini Pro (cloud fallback reasoning) Gemma 3 1B Gemma 3 1B (local general‑purpose model) Google DeepMind LFM 1.2B / 2.6B (Cactus lightweight foundation models) Python TypeScript Vite WebAssembly and LFM models) and confidence‑based evaluators custom routing controller local vector store structured logging

The full code is viewable in the first folder - the Cactus Github repository also got forked

Summarizing URL...