Claude Certified Architect - Professional study plan
Claude Certified Architect - Professional · Give each week a clear focus.
8 weeks · About 6 hours per week
Solution Architecture and Design
Map each sub-area of solution architecture (requirements decomposition, component boundaries, data flow, integration patterns) to a one-page decision checklist
Draft a reference architecture for a multi-step document-processing assistant, labeling every component and the interface between them
Identify three architecture trade-offs (latency vs. accuracy, coupling vs. reuse, cost vs. resilience) and write the criteria you would use to choose
Your checkpoint
A one-page annotated reference architecture diagram plus a written trade-off log covering at least three decisions
Solution Architecture and Design
Compare at least three orchestration topologies (single agent, supervisor/worker, pipeline) and state when each is the correct default
Design the failure and fallback path for each external dependency in your week-1 architecture
Write acceptance criteria for an architecture review that a peer could apply without you present
Your checkpoint
An architecture decision record set (minimum three ADRs) with context, options considered, decision, and consequences
Model Selection, Prompt Engineering, and Context Management
Build a model selection matrix that scores candidate models on task fit, context window, latency, cost per task, and tool-calling reliability
Rewrite one weak prompt into a structured prompt using explicit role, constraints, output schema, and few-shot examples
Design a context management strategy that specifies what is persisted, summarized, and evicted across a long session
Measure how output quality changes when you halve the available context budget
Your checkpoint
A prompt and context playbook containing the model selection matrix, two before/after prompt pairs, and a documented context budget policy
Model Selection, Prompt Engineering, and Context Management & Tool Use, Agentic Patterns, and Orchestration
Define JSON schemas for at least three tools with typed parameters, required fields, and explicit error contracts
Implement a supervisor/worker pattern where the supervisor delegates and validates worker output
Add retry, timeout, and idempotency handling to every tool call path
Document the stopping conditions that prevent an agent loop from running unbounded
Your checkpoint
A working reference implementation of a supervisor/worker agent with three schema-validated tools and explicit termination logic
Tool Use, Agentic Patterns, and Orchestration & Retrieval, Embeddings, and Knowledge Integration
Design a retrieval pipeline end to end: chunking strategy, embedding choice, index configuration, and reranking step
Justify chunk size and overlap choices against a specific document type rather than defaults
Implement hybrid retrieval combining keyword and vector search and compare results on ten queries
Write the grounding contract that tells the model how to cite retrieved sources and what to do when retrieval returns nothing relevant
Your checkpoint
A retrieval design document plus a ten-query comparison showing hybrid versus vector-only results with relevance judgments
Retrieval, Embeddings, and Knowledge Integration & Evaluation, Testing, and Quality Assurance
Build an evaluation set of at least thirty cases spanning typical, edge, and adversarial inputs
Define scoring rubrics with pass/fail thresholds for correctness, groundedness, and tool-use appropriateness
Automate a regression run that compares the current build against a stored baseline
Diagnose at least two observed failures and trace each to a root cause in prompt, retrieval, or tool design
Your checkpoint
A runnable evaluation suite with thirty-plus cases, scoring rubrics, and a baseline-versus-current regression report
Safety, Security, and Responsible AI Governance
Threat-model the system for prompt injection, data exfiltration, tool misuse, and over-permissioned credentials
Specify input and output guardrails and state where each is enforced in the request path
Define a least-privilege permission model for every tool and data source in the architecture
Write the human-in-the-loop escalation criteria for high-impact actions
Your checkpoint
A security and governance dossier containing a threat model, a guardrail placement map, and a least-privilege permission matrix
Deployment, Operations, Cost, and Performance Optimization & Evaluation, Testing, and Quality Assurance
Instrument the system with latency, token usage, and cost-per-request metrics broken down by pipeline stage
Identify the three highest-cost stages and propose a concrete optimization for each
Define the deployment rollout plan including canary criteria, rollback triggers, and monitoring alerts
Run a full timed review covering all seven domains and log every gap with a remediation action
Your checkpoint
An operations runbook with the cost and latency breakdown, three optimization proposals, a rollout and rollback plan, and a completed gap log
Milestones
Week 2: You can defend your reference architecture to a peer and they can restate your key trade-offs without prompting
Week 4: The supervisor/worker agent runs end to end, handles an injected tool failure gracefully, and terminates without manual intervention
Week 6: The evaluation suite runs unattended and produces a pass/fail regression report against a stored baseline
Week 8: You can complete a timed mock exam across all seven domains and every missed item has a written remediation action