Practical AI implementation and automation
The model is one component. Useful business AI also needs permissions, data, tools, monitoring and a definition of success.
Workflow selection
Good AI projects begin with a process that is repetitive, information-heavy or slow enough that improvement can be measured.
- Current process mapping
- Time/cost estimate
- Data sources
- Human approval points
- Risk review
- Success criteria
System design
The implementation may use one model or several, depending on the task.
- Model/API selection
- Retrieval and search
- Tool access
- Queues and scheduled work
- Structured outputs
- Logs and retries
- Human approval
Operational safety
Agents should not receive more authority than the job requires.
- Least privilege
- Approved tools
- Secret isolation
- Rate limits
- Auditability
- Failure handling
- Manual override
AI-assisted production stack
AI accelerates research, implementation, testing and iteration. Human engineering judgement remains responsible for the finished work.
Full text list
- OpenAI — Frontier models, Codex, agents, coding, research and business AI. Provider information
- Anthropic — Claude models and AI-assisted analysis, coding and workflow support. Provider information
- Google Gemini — Multimodal models, research, coding and agentic workflows. Provider information
- GitHub — Version control, code review, delivery history and controlled deployment. Provider information
- Ollama — Local and self-hosted model workflows where that architecture is useful. Provider information
- Hugging Face — Open models, datasets and specialist AI tooling. Provider information
Infrastructure, delivery and operations
Production sites can combine global edge services with conventional cloud/server infrastructure according to the workload.
Full text list
- Cloudflare — Global edge delivery, DNS, CDN, TLS, DDoS protection, WAF, rate limiting, Workers and storage services. Provider information
- Hetzner — Cloud and dedicated compute, networking, storage and conventional server workloads. Provider information
- Linux — Production server operating systems and secure service hosting. Provider information
- Docker — Portable containers for applications, services and controlled deployment. Provider information
- PostgreSQL — Production relational data storage for applications and services. Provider information
- GitHub — Source control, deployment history and auditable delivery workflows. Provider information
- Node.js — JavaScript runtime for APIs, tooling and server-side applications. Provider information
- Nginx — Web serving and reverse-proxy capability where conventional servers are appropriate. Provider information
Web standards, code and frameworks
We are not locked to a single framework. Straightforward sites should stay straightforward; applications can use deeper stacks when justified.
Full text list
- HTML5 — Semantic document structure and modern browser standards. Provider information
- CSS — Responsive layout, presentation, accessibility and performance-conscious styling. Provider information
- JavaScript — Standards-based browser and application behaviour. Provider information
- TypeScript — Typed application development for maintainability and safer change. Provider information
- React — Component-based user interfaces when an application needs them. Provider information
- Next.js — Full-stack React framework for suitable applications and content sites. Provider information
- Vue — Progressive user-interface development. Provider information
- Svelte — Compiled, performance-conscious user-interface development. Provider information
- Astro — Content-focused websites with low client-side JavaScript where appropriate. Provider information
- Python — Automation, data processing, APIs and supporting services. Provider information
The detailed pages exist for customers who want to inspect the engineering. The customer-facing offer remains deliberately straightforward.
