GCP GenAI Leader study plan

Define generative AI, large language models, and foundation models, and contrast them with classical ML and discriminative models

Explain tokens, embeddings, context windows, and the transformer attention mechanism at a conceptual level in your own words

Distinguish model modalities (text, image, speech, code, multimodal) and match each to a representative Google Cloud model family

List the main causes of hallucination and describe at least three mitigations at a conceptual level

Trace the generative AI model lifecycle from data preparation through training, adaptation, evaluation, and deployment

Compare prompting, grounding, and tuning on required data, cost, latency, and when each is the right first move

Derive the criteria used to select a model: quality, latency, cost per token, context length, modality, and compliance

Explain how evaluation works for generative tasks, including automated metrics and human/LLM-as-judge review

Map the Google Cloud GenAI portfolio: Vertex AI Studio, Model Garden, Gemini models, Imagen, and speech and embedding models

Distinguish the Gemini model tiers and select the appropriate tier for latency-sensitive versus quality-sensitive workloads

Describe the enterprise platform capabilities: VPC Service Controls, CMEK, data residency, and the commitment that customer data is not used to train foundation models

Practice prompt design in Vertex AI Studio and export a prompt to code

Explain how grounding with Vertex AI Search and the RAG Engine works end to end, from ingestion and chunking to retrieval and citation

Describe Model Garden deployment options: fully managed API, self-deployed endpoints, and tuning services

Identify agent-building capabilities and when an agent is preferable to a single-turn model call

Estimate cost drivers for a GenAI workload and identify the levers that control spend

Apply zero-shot, few-shot, role, and chain-of-thought prompting patterns to the same task and compare the outputs

Write prompts that specify output format, constraints, and length, and verify the model respects them

Diagnose failure modes such as vague instructions, conflicting constraints, and missing context, then rewrite the prompt to fix each

Build a small rubric and use it to score prompt versions objectively

Design a RAG pipeline for a document corpus, covering chunking strategy, retrieval quality, and citation of sources

Compare adapter tuning, distillation, and RLHF-style approaches on data requirements, cost, and expected gain

Decide when grounding is sufficient and when tuning is justified, using a written decision rule

Explain how to measure groundedness and detect unsupported claims in generated answers

Explain the Google Secure AI Framework pillars and map each to a concrete control in a GenAI deployment

Identify privacy, data governance, and IP considerations, including PII handling, data residency, and responsible use of customer data

Describe how to evaluate models and applications for harmful output, bias, and safety, and how to set up human review

Outline organizational enablers: executive sponsorship, a center of excellence, use-case prioritization, and change management

Build a phased adoption roadmap covering pilot, evaluation, governance sign-off, and scaled rollout

Quantify business value for a GenAI use case using time saved, quality lift, and cost per interaction

Consolidate weak areas from earlier weeks into a prioritized review list and close them

Complete a timed practice assessment and review every missed item to root cause