RESEARCH & ENGINEERING SERIES

Sovereign Labor AI & Taxonomy Intelligence

In-depth engineering deep-dives, empirical benchmark audits, and local demonstration guides for our domain-specialized European occupational classification models.

Hugging Face: mazafard/esco-gemma4Quantized GGUF: mazafard/esco-gemma4-GGUF

Part 1 · Research & BenchmarksAI & Sovereign Models

Sovereign Labor AI: Fine-Tuning Gemma 4 on 13,900+ European ESCO Competencies with 94.8% ISCO-08 Accuracy

Generic LLMs hallucinate invalid occupational codes up to 48.2% of the time. We fine-tuned Google Gemma 4 on 27,788 multi-resolution pairs across all 19 relational tables of ESCO v1.2.1, achieving 94.8% 4-digit ISCO-08 accuracy and 0.0% taxonomic hallucination.

ESCO v1.2.1ISCO-08Gemma 4QLoRAModal A100Data Sovereignty

Mohammadreza· Founder & CEO, DestinAI

•

September 11, 2026

•

8 min read

Read Article

Part 2 · Local Setup & Run GuideLocal Inference & Mac Demos

Step-by-Step Guide: Running esco-gemma4 for Demo Use on Mac (Ollama) and Fast Experimentation with Unsloth

Learn how to run esco-gemma4 locally for offline demos, executive presentations, and research prototyping on Apple Silicon Macs using Ollama and Unsloth—with 100% GDPR compliance and zero cloud cost.

OllamaApple SiliconMac AppUnslothGGUFDemo Use

Mohammadreza· Founder & CEO, DestinAI

•

September 11, 2026

•

6 min read

Read Article