Researching the next layer of intelligence.
YecoAI is an Italian AI company building AI-First vertical solutions and privacy-first systems for real-world deployment. We focus on model design, serving infrastructure, evaluation harnesses, and cognitive architectures — with a strong emphasis on transparency, reproducibility, and EU regulatory compliance.
We are backed by Invitalia, the Italian government's agency for business development, and part of the AWS Startup Program.
🔬 Website: yecoai.com · ✉️ Contact
- Research-first: rigorous evaluation at every step, with documented trade-offs and constraints.
- Transparent: reproducible methodologies, clear benchmarks, deterministic APIs.
- Privacy-first: EU-ready compliance from day one, consent-aware data strategies.
| Area | Description |
|---|---|
| AI Infrastructure | Low-latency pipelines, evaluation harnesses, robust serving systems that scale from prototype to production. |
| Cognitive Architectures | Exogenous layers (e.g., Anti-Loop) that stabilise agent cognition and enable deterministic autonomy. |
| Privacy & Compliance | Privacy-aware data strategies, GDPR-ready design, EU AI Act alignment. |
| Security Research | Prompt injection defence with structured frameworks, sandbox guards, and deterministic filters. |
An autonomous AI support agent for hosting providers. Integrates with WHMCS, Proxmox, and VirtFusion to automate L1 and L2 support tickets — 24/7, with multi-layer safety and GDPR compliance.
Currently in Beta.
A proprietary micro-LLM (278M parameters) for real-time PII masking. Anonymises names, IPs, passwords, IBANs, tax codes, and health data — on-device, with <50ms latency. GDPR and EU AI Act compliant.
Launching Q4 2026.
A 120M parameter Italian Mini PoC LLM. Outperforms Emma-5 550M on 5/5 benchmarks, trained on cloud GPU for <15€, CPU-optimised.
Experimental.
| Project | Description | Stack |
|---|---|---|
| YecoAI Cognitive Layer | Lightweight layer that detects loops, amnesia, and semantic degradation in LLM outputs (96% accuracy, 1.7ms latency). | Python |
| YecoAI Security Layer | Deterministic, ultra-low-latency security layer implementing Asimov's laws injection, context-aware N-gram filtering, Enterprise DLP, and sandbox guards. | Python |
| ArchGen AI | Generates interactive architecture diagrams from text, plus complete technical documentation, stack advice, security & cost insights. | TypeScript |
- DigithON 2026 Finalist - Top 100 startups in Italy
- Invitalia Grant - Funded by the Italian government
- AWS Startup Program - Cloud infrastructure partner
© 2026 YecoAI · Founded by Marco Nasi