Machine Learning & Deep Learning
We develop ML/DL models tailored to your specific business needs, from predictive analytics to computer vision and natural language processing.
Artificial Intelligence
Local AI, agentic AI and governed AI
Real-world governed-AI applications
We develop governed-AI projects for our clients with mature, verifiable tools: applications and agents that automate processes, gain insights from data and adapt to how the organisation works. Every model call goes through verifiable policies, personal data is redacted before it leaves the perimeter and every interaction leaves an audit trail.
Local AI, open-weight models
When data cannot leave, AI runs in-house: open-weight LLMs running locally, RAG on the organisation's documents and on-premise agents. Among the various platforms we adopt NVIDIA GB10, which we configure, harden and size to the client's workloads.
Agentic AI
We design agents and multi-agent systems with tool use, memory, MCP and A2A, on the leading agentic frameworks. Agents read, prepare and propose; actions with impact always require a person's approval.
noze adopts OISG (Open, Intelligent, Secure, Governed), the paradigm for autonomous AI systems proposed by Stefano Noferi: its four pillars guide the architecture, security and governance of every project.
We build them on the client's own systems and models, on-premise or in a private cloud, also starting from Admina, the open-source AI governance framework (Apache 2.0), a project by Stefano Noferi, sponsored by noze. Our products IntelliPA, DataGovern, VerdePilot and AgenticHealth are built on Admina too.
Large Language Models & Generative AI
We integrate and customize the main commercial and open-weight LLMs. We develop generative AI solutions for text, images, and code.
Retrieval-Augmented Generation (RAG) & Local LLMs
We implement RAG with local or cloud LLMs to integrate your proprietary data. We offer private solutions with open-weight models run locally, keeping data and control on-premise.
Agentic AI & Multi-Agent Systems
We design AI agents with reasoning, persistent memory and tool-use capabilities: they read, prepare and propose, while decisions stay with people. We build multi-agent systems with the leading agentic frameworks, implementing A2A architectures for interoperable agents.
Model Context Protocol (MCP)
We implement MCP to enhance AI capabilities by providing contextual awareness and access to external tools and data sources.
AI Ethics & Governance
AI governance following the OISG paradigm: verifiable policies, bias detection, transparency, audit trails and human approval of critical actions.
AI Integration
We integrate AI capabilities into your existing systems and workflows, getting the most out of the technology investments you've already made.
Custom AI Development
We build bespoke AI solutions tailored to your specific industry challenges, from clinical research to the public sector.
AI for Sensitive & Healthcare Data
We implement secure AI solutions for sensitive and healthcare data, with local models, encryption and controls that support regulatory compliance.
Our AI Methodologies
At noze, we employ a comprehensive approach to AI development, combining proven methodologies and recent techniques, chosen to fit the problem at hand.
Our AI expertise includes:
- Supervised, unsupervised, and reinforcement learning
- Neural networks and deep learning architectures
- Natural Language Processing (NLP) and Understanding (NLU)
- Computer Vision and image recognition
- Time series analysis and forecasting
- Anomaly detection and pattern recognition
- LLM Implementation & Fine-tuning (commercial and open-weight models)
- Agentic AI Frameworks: LangChain, LangGraph, CrewAI, AutoGen, LlamaIndex, smolagents, Google ADK, Cheshire Cat
We work closely with your team to understand your specific challenges and objectives, then develop AI solutions with agreed objectives and verification criteria. Our development process is iterative and includes continuous improvement and adaptation to changing business needs.
Frequently Asked Questions
What types of AI projects have you delivered?
We have delivered a wide range of AI projects: predictive analytics systems, custom enterprise chatbots with LLMs, computer vision systems for quality control, NLP solutions for document analysis, human-in-the-loop AI agents and multi-agent pipelines, and RAG systems with local and cloud models.
How do you integrate AI into existing systems?
Our approach to AI integration is modular and flexible. We start with a thorough analysis of existing systems, identify optimal integration points, and develop AI solutions that fit into existing workflows, minimizing operational disruption.
What AI technologies and frameworks do you use?
We work with a mix of well-established industry technologies: TensorFlow, PyTorch and Keras for deep learning; leading commercial and open-weight models for generative AI; the main agentic frameworks for multi-agent systems; and MCP (Model Context Protocol) for integration with external tools and data. The choice depends on specific project requirements.
How do you handle data privacy in AI projects?
Data privacy is a project requirement. We implement rigorous security measures, use data anonymization techniques, and follow industry best practices and privacy regulations like GDPR. We also offer on-premise AI solutions for particularly sensitive data.
How long does it take to implement an AI solution?
Implementation times vary based on project complexity. Typically, a proof-of-concept can be developed in 4-8 weeks, while a complete solution may take 3-6 months. We adopt an agile approach with incremental releases to deliver value from the early stages of the project.
Do you develop custom governed-AI projects?
Yes. We develop governed-AI projects for clients, with verifiable policies, redaction of personal data and an audit trail on every interaction with the models, on their own systems and models, on-premise or in a private cloud. We also start from Admina, the open-source AI governance framework (Apache 2.0), a project by Stefano Noferi, sponsored by noze.
Do your AI agents act autonomously?
No: in our systems, agents read, prepare and propose, but any action with real impact (sending, signing, making changes) always requires a person's approval.
Related products
Product
CyberAgent
The AI agent that watches your attack surface, validates what is actually exploitable and prepares NIS2 evidence.
ProductIntelliPA
Turnkey self-hosted AI for Italian municipalities and public bodies.
ProductDataGovern
AI agents that oversee your documents and business processes and prepare NIS2, GDPR and AI Act compliance.
ProductAgenticHealth
On-premise agentic AI for clinical research and quality at healthcare providers. Not medical software.
ProductVerdePilot
The AI copilot for environmental and energy compliance at sites in Italy.
Designed for
Industry
SMEs
NIS2, security and AI governance for SMEs.
IndustryPublic Administration
On-premise AI for municipalities, provinces and regions.
IndustryHealthcare
On-premise AI for clinical research, with data kept at the provider.
IndustryEnvironment
Environmental and energy filings with verifiable drafts.
IndustryProfessional Firms & Consultants
NIS2, AI Act and GDPR compliance for firms and consultants.
IndustryMSPs & SIs
White-label, enterprise support, dedicated channel.
IndustryResearch & Academia
25+ years of applied R&D, partner of European consortia.
Contact
Want to dig deeper?
Fill in the form: your request reaches the dedicated team.