Technical architecture
How our solutions are built.
A layered architecture with open, replaceable components, designed to integrate with the systems the company already uses and to run where the data has to stay.
Layers · sample
- Interface: web dashboard and API
- Models: on-premise LLM
- Data: local vector store
The five layers
From interface to infrastructure.
Interface
Web dashboards, chat and REST APIs to integrate AI functions into the tools the team uses every day.
Application
Containerised microservices orchestrating models, business rules and workflows.
Models
Language models, computer vision and language processing, hosted on-premise or in a private cloud.
Data
Relational and vector databases, message queues and pipelines for structured and unstructured data.
Infrastructure
On-premise servers, European private cloud or hybrid architectures, with infrastructure as code.
Integrations
MCP connectors to ERP, CRM and document archives.
Technology stack
Open, replaceable components, no vendor lock-in.
Models and orchestration
- Open and commercial language models (OpenAI, Anthropic, Mistral, Meta)
- PyTorch, Hugging Face, vLLM for local execution
- LangChain and MCP servers for orchestration
- MLflow for the model lifecycle
Data and application
- PostgreSQL, Redis, vector databases
- Apache Kafka for real-time flows
- FastAPI and Kubernetes for services
- Docker and Terraform for repeatable releases
Security and monitoring
- Encryption at rest and in transit
- Role-based access control, single sign-on
- Activity log for GDPR audits
- Prometheus and Grafana for monitoring
Security
Data protection built into the design.
Data protection
Encryption, masking and secure key management.
Privacy by design
Data minimisation, consent management and right to erasure, as required by GDPR.
Access control
Role-based permissions, multi-factor authentication and a full activity log.
Compliance
GDPR and EU AI Act; support for audits by the client and its advisers.
Integrations
Systems our solutions talk to.
| Category | Systems | Method |
|---|---|---|
| ERP | Zucchetti, TeamSystem, SAP, Odoo | MCP connectors and APIs |
| CRM | Salesforce, HubSpot, Zoho, Pipedrive | MCP connectors and APIs |
| Collaboration | Microsoft 365, Google Workspace, Slack, Teams | APIs and webhooks |
| Support | Zendesk, Freshdesk, ServiceNow | APIs and webhooks |
Frequently asked questions
Do the solutions work without the cloud?
Yes. Models, data and application can run entirely on servers in the company. The European private cloud is an option, not a requirement.
Can I switch language model later?
Yes. The architecture separates models from the application: you can move from an open model to a commercial one, or vice versa, without rewriting the integrations.
How are updates managed?
With controlled, repeatable releases (infrastructure as code), automated tests and the ability to roll back to the previous version.
Let’s talk about your business.
45 minutes with a PugliAI consultant: process review, priority opportunities and a written report with an ROI estimate. Free and without obligation.