Kursplan
Foundations of Architecture for AI Systems
- Defining layers, boundaries, and the separation of concerns
- Applying ports and adapters patterns to AI components
- Establishing quality attribute scenarios for AI features
- Utilizing Architecture Decision Records as a governance tool
Modernizing Existing Applications
- Implementing incremental modernization rather than full rewrites
- Positioning AI services alongside existing monolithic structures
- Employing Strangler and Anti-Corruption Layer patterns
- Creating integration points for existing APIs and data structures
Architectural Patterns for LLM Integration
- Using the model gateway as an abstraction boundary
- Implementing synchronous, asynchronous, and streaming integrations
- Managing fallbacks, degradation, and provider failover strategies
- Treating prompt and model configurations as version-controlled assets
Contracts, APIs, and Data Exchange
- Defining structured output schemas and validation rules
- Implementing idempotency, retry mechanisms, and rate limiting
- Managing latency budgets and timeout orchestration
- Considering cost per request as a key architectural constraint
End-to-End RAG Architecture
- Designing ingestion, chunking, and embedding pipelines
- Structuring index, retrieval, reranking, and generation stages
- Determining retrieval placement relative to application services
- Analyzing trade-offs and potential failure modes at each stage
Vector Databases and Retrieval Design
- Evaluating managed versus self-hosted vector stores
- Implementing dense, sparse, and hybrid search techniques
- Managing metadata filtering and tenant isolation
- Planning re-embedding strategies for model changes
Retrieval Quality, Grounding, and Evaluation
- Constructing evaluation sets based on real-world queries
- Measuring retrieval and answer quality metrics
- Managing citations, grounding, and hallucination containment
- Incorporating feedback loops into the indexing process
Agent Architectures
- Understanding the agent loop and tool invocation mechanisms
- Comparing deterministic workflows with autonomous planners
- Identifying scenarios where an agent is not the appropriate solution
- Defining the scope of agent responsibilities and state management
Multi-Agent Systems and Orchestration
- Implementing supervisor and orchestrator topologies
- Defining handoff protocols and message contracts
- Managing shared memory and context
- Comparing frameworks based on architectural criteria
Service Communication and Tool Protocols
- Using event-driven and queue-based coordination methods
- Exposing tools through standard protocols
- Managing sagas, compensation, and long-running transactions
- Ensuring observability across multiple service hops
AI Agents in Development Workflows
- Integrating coding agents into the delivery pipeline
- Adopting spec-driven development and review gates
- Implementing test generation and automated quality checks
- Establishing guardrails to maintain architectural control
Governance and Software Quality
- Conducting design reviews for AI features
- Managing model and prompt changes
- Including regression evaluation as a pipeline stage
- Defining ownership, policies, and approval workflows
Security, Privacy, and Data Governance
- Addressing prompt injection and defense-in-depth strategies
- Managing secrets, access control, and least privilege principles
- Handling data residency, personal data, and retention policies
- Ensuring auditability and human-in-the-loop controls
Scalability, Cost, and Observability
- Tracing processes across agent chains
- Monitoring token usage and cost telemetry
- Optimizing caching and latency
- Defining service levels and incident response for AI services
Capstone: Architecture for an AI-Driven Solution
- Developing context, component, and deployment views
- Designing retrieval and agent orchestration systems
- Creating a security, governance, and cost management plan
- Conducting peer reviews, critiques, and roadmap planning for migration
Krav
- Erfaring med design og vedlikehold av produksjonsprogramvare
- God kjennskap til REST-APIer og tjenesteintegrasjon
- Kunnskap om relasjonelle databaser og datamodellering
- Grunnleggende kjennskap til LLM-baserte verktøy eller API-er
Publikum
- Programvarearkitekter
- Tekniske ledere og senior ingeniører
- Utviklingsledere med ansvar for arkitekturvalg
Referanser (1)
At jeg har skaffet kunnskap om Streamlit-biblioteket fra Python, og jeg vil sikkert prøve å bruke det for å forbedre applikasjonene i min gruppe som er laget i R Shiny.
Michal Maj - XL Catlin Services SE (AXA XL)
Kurs - GitHub Copilot for Developers
Maskinoversatt