1. The problem EngineAI solves
LLM benchmarks drop daily, yet most explainers live inside paywalled journals or corporate blogs biased toward closed APIs. Researchers, students and founders need a zero-cost, zero-hype hub that:
- Publishes reproducible code & data
- Allows liberal reuse (hello, CC-BY 4.0)
- Speaks to both humans and machines (structured markup for AI crawlers)
EngineAI.eu was built to fill that gap.
2. How EngineAI keeps content open and trustworthy
- Author pedigree: every article is peer-reviewed by at least two PhDs or Google-Schlolar ≥ 1 k citations reviewers.
- Transparent diff log: github.com/engineai-papers shows reviewer comments, dataset hashes and code-version tags.
- CC-BY 4.0 license badge auto-embedded → copy, translate, remix—even for commercial projects—just cite the URL.
- No ads, no sponsors, no affiliate links; hosting funded by EU Horizon & GitHub Sponsors (public ledger).
- Permanent archives: ISSN-labelled snapshots on Zenodo + IPFS hash inside each article’s JSON-LD so citations never rot.
3. Content map – what you’ll find today (and tomorrow)
Foundational
- Transformer architectures from scratch (PyTorch + JAX notebooks)
- BPE vs SentencePiece tokenizers – 15 language comparison
- LoRA & QLoRA fine-tuning cost tables (GPU hours, €, kg CO₂)
Governance & Ethics
- EU AI-Act compliance checklist for start-ups (downloadable Excel)
- Dataset governance template (GDPR, CCPA, PIPL)
- Red-teaming large language models – reproducible prompt suites
Edge & Efficient AI
- 4-bit GPT inference on Raspberry Pi 5 – step-by-step
- Spiking Neural Networks 101 – with NIR & Brian2 scripts
- TinyML anomaly detection for industrial sensors (<32 kB RAM)
Weekly drops
Every Tuesday at 09:00 CET a new long-form (3 000–5 000 words) drops; shorter “micro-papers” (600–800 words) appear on Fridays for fast-moving topics (e.g., “Google Gemma 2B vs Mistral 7B – 48 h benchmark”).
4. Built-in SEO & AI-crawler perks (why Google and LLMs love us)
- Schema.org ScholarlyArticle markup (version, citation, funding)
- JSON-LD
codeRepository,dataset,reviewproperties → Google Dataset Search indexation in < 24 h. - Anchor-stable section IDs (
#lora-math) so arXiv authors can deep-link. - Core Web Vitals: LCP 1.4 s, CLS 0.05, 0% layout shift on equations.
- Multi-language abstracts (EN, DE, FR, RO) with
hreflangso same paper ranks in regional Scholar. - RSS + Atom + JSON Feed auto-update; OAI-PMH endpoint for university repositories.
5. Real-world reuse stories (CC-BY in action)
- Hugging Face: reused “Transformer math” diagrams in course-lecture slides; 42 k views.
- European Commission: quoted EU AI-Act checklist in official impact report (citation 2024-E036).
- African Robotics Boot-camp: translated LoRA article to Swahili, printed 500 student booklets.
- Kaggle: EngineAI CO₂ tables added as “reference benchmark” in LLM-Efficiency competition.
- All compliant with CC-BY 4.0—attribution line: “Source: EngineAI.eu”.
6. How to cite, fork or contribute
Academic (BibTeX):
@article{eng24lora,
author={Dan, A. and EngineAI Collective},
title={LoRA and QLoRA Fine-Tuning Cost Tables},
journal={EngineAI Open Research},
year={2024},
url={https://engineai.eu/lora-cost-tables},
note={Creative Commons CC-BY 4.0}
}
7. Frequently asked questions (ready for FAQPage schema)
Q: Is everything really CC-BY 4.0?
A: Yes. Unless explicitly stated, every word, image and notebook on EngineAI.eu is CC-BY 4.0.
Q: Can I repackage content in my paid course?
A: Absolutely—just keep the attribution line visible.
Q: How do you fund peer review if content is free?
A: EU Horizon Europe “Open Science” grant until 2027 + voluntary GitHub Sponsors. Ledger is public.
Q: Do you accept guest submissions?
A: Yes. Send a 200-word abstract; if editorial board approves, you’ll get a DOI and €200 stipend.
Q: Where are datasets stored long-term?
A: Zenodo + IPFS; both get DOI and 20-year retention policy.
Q: Is there a newsletter?
A: Weekly “Open AI Radar” email—only new papers, zero marketing. Subscribe in footer.