AWS Certified Generative AI Developer (Professional) — my study path
AWS Generative AI Certification Bedrock
I passed the AWS Certified Generative AI Developer — Professional certification (and held the Early Adopter badge from the earlier track). This is the study path that worked for me, organised the way the exam actually tests you.
What the exam really measures
It is not a trivia quiz about model names. The Professional tier expects you to design and operate GenAI solutions: pick patterns, put them in production, evaluate them, and control cost and safety. Every question had a "what would you do first" flavour.
The four blocks I studied
1. Model access and selection on Bedrock
Know the trade-off matrix cold: latency vs. quality, on-demand vs. provisioned throughput, and when not to use the largest model. Multi-model strategy (route cheap requests to small models) showed up repeatedly.
2. RAG architecture
The core pattern. Chunking strategies, embedding model choice, hybrid search (semantic + keyword), reranking, and metadata filtering. Be able to diagnose a RAG pipeline that returns technically-similar but factually-wrong passages — usually a chunking or ranking problem, not a model problem.
3. Evaluation and safety
Offline vs. online evaluation, building a golden dataset, guardrails, prompt-injection defences, and content filtering. This block carried the most "production readiness" questions — the stuff that separates a demo from a shipped feature.
4. Cost and scaling
Token accounting, caching repeated context, right-sizing models, and monitoring with CloudWatch. I actually built a small usage-tracking dashboard for a client project, and the cost questions were much easier after that.
How I studied
- Two weeks, ~1 hour/day. Consistency beat cramming.
- Hands-on over videos. Every concept got a tiny Bedrock experiment — a RAG notebook, a guardrail test, a cost calculation.
- Practice exams for pacing only. They teach you to read the question stem carefully, not the content.
- Built projects with the patterns. Applying retrieval and evaluation to a real portfolio project cemented far more than any slide deck.
Would I recommend it?
If you are already shipping LLM features and want a structured, AWS-flavoured mental model for production GenAI — yes. If you have never called an API with a prompt, start smaller first.
The certification is a checkpoint. The actual skill is knowing why a pipeline behaves the way it does.