Enterprise AI architecture
Infrastructure to applications, governed at every layer.
Infrastructure → Data → Model Training → Model Serving → RAG → Agent Orchestration → APIs → Enterprise Applications → Monitoring.
Interactive architecture
Request path
Layer 1 of 9
Infrastructure
Compute, networking and container platform.
Key controls
- Private networking
- Containerisation
- Infrastructure as code
Engineering topics
What the architecture covers
- Deployment models
- On-premise, private cloud, dedicated or hybrid — chosen per workload and data classification.
- Model hosting
- Open-weight and fine-tuned models served privately; external models used only where approved.
- Private networking
- Services communicate on private networks; public exposure only through the gateway.
- API gateway
- Single entry point for authentication, quotas, routing and logging.
- Identity & access
- SSO integration, role-based access and service identities for agents.
- Secrets management
- Credentials kept in a secrets store, never in code or prompts.
- Data encryption
- TLS in transit and encryption at rest for storage that supports it.
- Audit logging
- Prompts, retrievals, tool calls and approvals logged for review.
- Containerisation
- Services packaged as containers for portable, repeatable deployment.
- CI/CD
- Automated build, test and deployment pipelines with review gates.
- Observability
- Metrics, logs and traces for models, retrieval and agents.
- Business continuity
- Backups, restore testing and documented recovery procedures.
- Responsible AI
- Evaluation before release, human oversight for consequential actions, and clear disclosure of AI use.
This reference architecture describes ADGTech's design approach. The specific technologies, hosting providers and controls in a deployment are documented in its architecture specification.
Get an architecture tailored to your environment
Talk to ADGTech engineers about models, knowledge, agents and where your AI should run.