Solution Patterns
Production AI solutions are typically built as modular architectures rather than a single model or component. The foundation model provides core reasoning and generation capabilities, while retrieval connects AI to trusted enterprise knowledge, agents determine actions and use tools, and orchestration coordinates workflows across models, agents and services. Protocols such as MCP can provide standardised access to external tools and data. Guardrails, security, evaluation, monitoring and governance should be treated as cross-cutting controls, helping ensure AI solutions operate safely and reliably as they move from experimentation into production.
How These Pieces Fit Together

It reads left to right: sources feed ingestion, the medallion layers carry the data from raw landing zone to a curated mart, business context (semantics, metadata, ontology, knowledge graph) makes it interpretable, and the knowledge layer holds retrieval storage alongside session state and long-term memory. The runtime column is what consumers actually reach through: a models gateway, MCP and APIs, and an agent framework. Governance and orchestration are drawn as bands underneath because they apply across every layer rather than occupying one point in the flow.
The running question, "What was revenue last quarter by region?", is built in two patterns on this list. Text-to-SQL turns the question into a validated query, and The AI Layer supplies the governed definitions that make the answer trustworthy. AI Security, Safety & Governance covers the controls that keep it read-only.
In This Section
The patterns fall into five groups. Read them in order: each group builds on the one before it.
AI Data & Knowledge
Getting enterprise documents and data in front of a model.
- Intelligent Document Processing - OCR, layout detection, table extraction, and LLM-powered structured extraction with confidence scoring.
- RAG & Enterprise Knowledge - chunking strategies, retrieval architectures, re-ranking, evaluation, and failure modes.
- Conversational Data & Analytics - asking questions of structured data in plain language:
- Text-to-SQL - schema grounding, query generation, validation, and production deployment.
- The AI Layer - semantic definitions, verified examples, and metric trees that make AI-powered self-service analytics trustworthy.
Agentic AI
Models that choose actions and call tools.
- AI Agents - tool use, function calling, the ReAct pattern, memory systems, and deployment patterns.
- MCP & Tool Integration - building MCP servers, connecting tools to LLMs, and transport mechanisms.
- Multi-Agent Systems - sequential, concurrent, handoff, group chat, and magentic orchestration.
AI Automation
Running agents inside business processes, with or without a person in the loop.
- Autonomous AI - levels of autonomy, human-in-the-loop patterns, and the path from supervised to self-directed agents.
- AI Workflow Automation - triggers, orchestration, task pipelines, and the operational scaffolding around them.
AI Models & Engineering
Choosing, adapting, and operating the model itself.
- Model Strategy - when to fine-tune versus prompt engineering or RAG: trade-offs, workflows, and cost.
- AI Engineering / LLMOps - evaluation, deployment, and monitoring of LLM applications. Planned for a later release.
Trusted AI
Controls that apply across every group above.
- AI Security, Safety & Governance - content filtering, prompt injection defence, PII protection, and responsible AI patterns.
Next: Intelligent Document Processing - every pattern starts with getting source material into a form a model can use.