Decentralized Inference Mesh (Google Cloud Hackathon)
February 2026 — Google Cloud Hackathon
The first EU AI Act-native orchestration layer for privacy-critical AI applications, enabling compliant AI deployment using local Small Language Models (SLMs) without cloud data egress.
Key Insights
- Developed a Zero-Egress, governance-native orchestration layer for local inference using Ollama.
- Implemented an automated regulatory classification engine mapping projects to EU AI Act articles (Art. 5, Annex III, Art. 50).
- Engineered a “Trust & Safety Shield” for real-time PII blocking and prompt injection prevention.
- Built a persistent compliance vault for automated, immutable audit trails of agent interactions (Art. 12).
- Designed a sequential Multi-Agent Hive architecture decomposing tasks into specialized Architect, Frontend, and Backend agents.
Tech Stack
- Python (FastAPI, Pydantic State Schema)
- Ollama (Local SLMs: qwen2.5-coder, llama3)
- Streamlit (Visual Dashboard)
- Docker & Google Cloud (Vertex AI Hybrid concepts)