All courses
Community Incubator Engineering & Strategy track
Free
advanced
AI Architecture
Design AI systems that are useful, safe, and honest about their limits. You will frame problems, choose between building and buying, design RAG and agent systems, and plan for evaluation, security, privacy, cost, and failure. Every module asks when AI is the wrong answer. Free, like every Community Incubator course.
- Track
- Engineering & Strategy
- Length
- 12 weeks
- Live sessions
- Weekday evenings
- Modules
- 11
- Level
- advanced
- Cost
- Free
Outcome
Design, document, and defend an AI system architecture with an evaluation plan, a risk register, and clear lines for when AI should not be used.
Capstone
Write an architecture document, risk register, and evaluation plan for an AI assistant at a real community organization.
Focus areas covered
1 of 5 in the Engineering & Strategy track.
- AI Architecture
Modules
- Module 1: Problem Framing and When Not to Use AITurn a vague request into a testable problem statement, and learn to say no to AI when a rule, a form, or a person is the better system.
- Module 2: Build vs Buy vs BlendCompare hosted APIs, open models, and managed services on data control, team fit, quality, and exit cost, then make a written decision you can defend.
- Module 3: LLM Application PatternsTreat prompts as versioned interfaces, force model output into validated structures, and let models call tools only through an allowlist you control.
- Module 4: Retrieval-Augmented GenerationBuild the retrieval half of RAG on purpose: chunk documents, embed and store them, cite sources, and measure whether retrieval actually finds the right passage.
- Module 5: Agents and OrchestrationDecide when a fixed workflow beats an agent, connect tools through MCP servers with tight scopes, and put human approval gates in front of every action that matters.
- Module 6: Data Pipelines and GovernanceDesign data flows that respect consent, track where every record came from, and handle personal information with extra care for minors.
- Module 7: Evaluation and GuardrailsBuild test sets before you build features, use LLM-as-judge only after checking it against people, red-team your own system, and layer safety filters with clear fallbacks.
- Module 8: Security and PrivacyDefend against prompt injection and data exfiltration, keep secrets out of prompts and code, and give every model and tool the least privilege it needs.
- Module 9: Cost, Latency, and ReliabilityMake AI systems fast and dependable under real load with caching, batching, model routing, retries, and fallbacks that degrade with care.
- Module 10: Deployment and ObservabilityShip AI changes in stages, log what you need without logging what you must not, watch quality as well as uptime, and respond to incidents with a plan.
- Module 11: Capstone: An Architecture That Earns TrustDesign a real AI system for a community organization: an architecture document, a risk register, and an evaluation plan, judged first on judgment.