T5: AI Architect
Track: AI Architect
Agentic stack planes: UX, Control, Runtime, Context & Memory, Data, Integrations, Security, Observability
Enterprise planes: People & Leadership, Culture & Change, Governance & Risk, Strategy & Portfolio, Economics & ROI
Target Audience: Senior architects and principal engineers
Badge: Applied AI Architect
Meaning: Intelligence ceiling controller
Agentic plane colors match the AI Capabilities Stack (round dots on track cards).
Enterprise plane colors match the same diagram (square markers on track cards).
Who This Is For
- Senior AI architects and system designers
- Principal engineers designing AI infrastructure
- Technical leads building advanced AI systems
- Architects working with memory, RAG, and full-stack agentic systems
What You'll Learn
Core Capabilities
- Temporal Memory - Design systems that learn and adapt over time
- Adaptive RAG - Build retrieval-augmented generation systems that evolve
- Context Orchestration - Manage complex context across systems
- Intelligence Ceiling Control - Design systems with appropriate intelligence boundaries
Key Focus Areas
- Memory vs retrieval: cognition-oriented memory networks
- Data plane: multi-tenant schemas and behavior contracts
- Full-stack synthesis and protocol ecosystems (MCP, A2A)
- Explainability-first context (for public sector)
Schedule
Duration: 12 weeks
Weeks: 1-12
Focus: Full stack - UX through observability, with depth on memory and data
Weeks 1-4: Foundations (AI Engineer track scope)
These weeks mirror the AI Engineer track foundation so architecture work rests on the same production baseline: architecture overview, control plane and tool registry, policy and guardrails, identity and zero-trust.
Week 1: Architecture Overview
Reference model, plane boundaries, failure modes, and target stack sketch - same themes as T4 Week 1.
Week 2: Control Plane + Tool Registry
MCP, tool registry, allowlists, and deterministic policy surfaces - same themes as T4 Week 2.
Week 3: Policy + Guardrails
Input/output pipelines, CI testing for guardrails, escalation patterns - same themes as T4 Week 3.
Week 4: Identity + Zero-Trust
Workload identity, OAuth patterns, secrets, and threat modeling for agentic flows - same themes as T4 Week 4.
Cognitive Integration: Completes shared foundations before deep-plane work.
Week 5: Context & Memory Plane
Bloom Level: Apply (primary), Create (secondary), Analyze (foundation)
Knowledge Dimension: Conceptual & Procedural
Tool Mediation: Vector stores, memory libraries, graph/temporal patterns
Vectors vs graphs (explicit): Embedding / vector retrieval supports similarity search over chunks (the usual RAG path). Graph databases hold relational structure (entities, edges) for multi-hop reasoning, constraints, and explainability. Many production designs use hybrid patterns - e.g. vector retrieval for candidates, graph traversal or joins for grounding - where both fit the problem.
- Memory as cognition - not “RAG only” [Understand]
- World, experience, opinion, observation networks (framing) [Analyze]
- TEMPR (Temporal Entity Memory Priming Retrieval) with Hindsight [Apply]
- Designing for update, decay, and consent [Create]
Cognitive Integration: Memory plane depth - Judgment on what to remember.
Week 5 artifact (assessment): A vendor-neutral memory-boundary spec—scoped principals, bank layout, recall strategy classes, update/decay/consent rules, and how the harness invokes memory vs how context is composed for the model. Optional implementation appendix: one example stack (e.g. Astrocytes + pgvector) with diagram references to the Glossary: Astrocytes (framework).
Week 6: Data Plane
Bloom Level: Apply (primary), Create (secondary), Analyze (foundation)
Knowledge Dimension: Procedural & Conceptual
Tool Mediation: Postgres, object storage, migrations, contracts
- Multi-tenant data ownership [Apply]
- Behavior schemas and tool I/O contracts [Create]
- Lineage and PII boundaries [Analyze]
- Schema as source of truth for agent behavior [Evaluate]
Cognitive Integration: Data plane as behavioral contract - Governance.
Week 7: Security Plane
Bloom Level: Apply (primary), Analyze (secondary), Evaluate (foundation)
Knowledge Dimension: Procedural & Metacognitive
Tool Mediation: mTLS, vaults, guardrail services, compliance refs
- Deepening identity and authorization across planes [Apply]
- I/O boundaries and exfiltration risk [Analyze]
- Compliance overlays (e.g. regional AI rules) as context [Understand]
- Defense in depth for agentic systems [Create]
Cognitive Integration: Security plane architecture - Rigor.
Week 8: Observability Plane
Bloom Level: Create (primary), Apply (secondary), Evaluate (foundation)
Knowledge Dimension: Procedural & Metacognitive
Tool Mediation: Full OTel-style stacks, LLM observability, eval platforms
- Full-stack tracing across tools, models, memory, and data [Create]
- SLOs for agentic workflows [Evaluate]
- Owner dashboards and escalation [Apply]
- Vendor portability [Analyze]
Cognitive Integration: Observability plane at architect level.
Week 9: Integrations + Protocols
Bloom Level: Create (primary), Apply (secondary), Analyze (foundation)
Knowledge Dimension: Procedural & Conceptual
Tool Mediation: MCP ecosystem, A2A agent cards, search/scrape APIs
- Protocol convergence: MCP, A2A, and related patterns [Understand]
- Drift and versioning for external tools [Analyze]
- Integration test strategy [Create]
- Selecting build vs buy for connectors [Evaluate]
Cognitive Integration: Integrations plane - Transfer across vendors.
Week 10: Full Stack Synthesis
Bloom Level: Evaluate (primary), Create (secondary), Analyze (foundation)
Knowledge Dimension: Metacognitive & Conceptual
- End-to-end walkthrough of requests across planes [Create]
- Tradeoff analysis (cost, latency, risk) [Evaluate]
- Single coherent reference architecture [Create]
Cognitive Integration: Holistic Judgment.
Week 11: Astrocytes-augmented architecture
Bloom Level: Create (primary), Evaluate (secondary), Analyze (foundation)
Knowledge Dimension: Metacognitive & Procedural
Using the open-source Astrocytes memory framework as a concrete governed-memory boundary (see Glossary: Astrocytes (framework))—not the biological metaphor alone.
- Advanced patterns for memory-augmented and layered cognition [Create]
- When augmentation helps vs adds fragility [Evaluate]
- Research-to-production guardrails [Analyze]
Cognitive Integration: Pushes Create and Evaluate at architecture altitude.
Week 12: Architecture Capstone
Bloom Level: Create (primary), Evaluate (secondary), Analyze (foundation)
Knowledge Dimension: Metacognitive & Procedural
- Final architecture documentation [Create]
- Annotated stack diagram (planes, risks, choices) [Create]
- Defense and credentialing [Evaluate]
- Capstone: Production architecture doc + annotated stack diagram [Create]
Cognitive Integration: Capstone artifact across dimensions.
Assessment: Production architecture doc + annotated stack diagram
Track Learning Outcomes
Upon completing this track, you will be able to:
- Design memory architectures - Create systems that learn, remember, and adapt over time
- Build adaptive RAG systems - Implement retrieval-augmented generation systems that evolve and improve
- Orchestrate context - Manage complex context across multiple systems, agents, and interactions
- Control intelligence boundaries - Design systems with appropriate intelligence ceilings and boundaries
- Optimize context efficiency - Balance context richness with performance and cost considerations
- Design explainable context - Create context systems that are transparent and auditable (especially for public sector)
- Architect end-to-end AI systems - Design complete AI systems integrating multiple components and capabilities
- Defend architecture decisions - Articulate and justify architectural choices to technical and non-technical stakeholders
Transfer & Judgment Capabilities
This track explicitly develops two critical capabilities:
Transfer Capability
Apply learning to novel situations and contexts
- Adapt memory architectures to different system requirements and constraints
- Transfer RAG patterns across various domains and use cases
- Generalize context orchestration to diverse system architectures
- Apply intelligence boundary principles to different capability levels
How it's developed: Through designing systems for different contexts, adapting architectures to new requirements, and applying patterns across domains.
Judgment Capability
Make sound decisions under uncertainty
- Exercise restraint in system capability and intelligence decisions
- Identify and avoid unsuitable memory or RAG approaches
- Make principled tradeoffs between context richness and efficiency
- Balance system capabilities with safety, cost, and governance requirements
How it's developed: Through architecture decision scenarios, "when NOT to use advanced memory" exercises, efficiency tradeoff analysis, and intelligence boundary frameworks.
Track Structure
This track contains multiple levels, each with specific learning outcomes. Progress through levels to build your capability as an AI architect.
Level 1: Foundations (Weeks 1-4)
Aligned with AI Engineer track scope: architecture overview, control plane and tool registry, policy and guardrails, identity and zero-trust - so advanced architecture rests on the same production baseline.
Level 2: Deep planes (Weeks 5-8)
Bloom Focus: Apply & Analyze (primary), Create (secondary)
Knowledge Dimensions: Procedural & Conceptual
- Context and memory plane; TEMPR (Temporal Entity Memory Priming Retrieval) with Hindsight; vectors vs graph stores and hybrid patterns [Create]
- Data plane: multi-tenant data, behavior schemas [Apply]
- Security plane depth [Apply]
- Full observability stack [Create]
Level 3: Synthesis (Weeks 9-12)
Bloom Focus: Evaluate & Create (primary), Analyze (foundation)
Knowledge Dimensions: Metacognitive & Procedural
- Integrations and protocol ecosystems (MCP, A2A) [Create]
- Full stack synthesis [Evaluate]
- Astrocytes-augmented (framework) and advanced architecture patterns [Create]
- Architecture capstone and defense [Evaluate]
Assessment
- Artifact: Production architecture doc + annotated stack diagram
- Focus: All dimensions - judgment, rigor, transfer, and governance
- Outcome: Ability to architect intelligent systems with controlled intelligence ceilings
Public Sector Emphasis
For public sector participants, this track emphasizes:
- Explainability-first context design
- Full audit trails
- Transparency in memory and context
- Accountability in system behavior
Credential
Upon completion, earn the Applied AI Architect badge - demonstrating your capability as an intelligence ceiling controller.
Ready to get started? Get in touch to learn more or apply.
Previous: T4: AI Engineer