README / README.md
Kebire's picture
Add Lab reports section
0a1f867 verified
|
Raw History Blame Contribute Delete
1.45 kB
metadata
title: README
emoji: 🏆
colorFrom: purple
colorTo: indigo
sdk: static
pinned: false

Strategic Center for Artificial Intelligence (SCAI)

SCAI is a non-profit organization based in Sofia, Bulgaria. We help small and medium-sized enterprises and public organizations in the CEE region adopt AI in a practical, responsible way: AI maturity assessments, EU AI Act and GDPR readiness, and hands-on guidance on open models, MLOps and AIOps. We develop concepts, technology adoption frameworks and experiments for sovereign generative AI.

Our AI lab

We test before we advise. Our lab runs open models on our own NVIDIA GB10 hardware with vLLM, Unsloth and TRL. We fine-tune, measure against our own baselines, and publish what we learn, including the experiments that failed.

Current focus areas:

  • Fine-tuning with LoRA, QLoRA and preference optimization (DPO)
  • Tool-use and agent behavior: when a model should call a tool, ask, or decline
  • Evaluation design for small and mid-sized open models
  • Agentic workflows on the NVIDIA Nemotron 3 model family and other open models

Lab reports

  • Fine-Tuning Is an Art (Block B, September 2026). LoRA, QLoRA and DPO on a 4B open model: a real gain on tool-use decisions, a regression on coding, and the rules we now apply before fine-tuning. Read the article

Contact

Website: scaieurope.com