ACP-220 study plan

Map the ACP-220 domain weights to a personal study calendar and allocate weekly hours accordingly

Define foundational generative-AI terms (foundation model, token, embedding, inference, fine-tuning, RAG) in your own words

Outline the PAI model lifecycle stages from data preparation through deployment and monitoring

Compare at least three model families by typical use case, input modality, and cost/latency tradeoffs

Explain how training, validation, and evaluation splits differ and why leakage matters

Describe common evaluation metrics for classification, generation, and retrieval tasks

Identify which lifecycle stage each of five sample project scenarios is in

List the artifacts (datasets, checkpoints, configs, eval reports) that should be versioned at each stage

Sketch the architecture of a generative AI application: client, orchestration, model endpoint, retrieval store, guardrails

Describe how prompt templates, system instructions, and few-shot examples affect output quality

Explain the role of embeddings and vector search in retrieval-augmented generation

Draft a design for a document Q&A app including data flow and failure points

Build a working prototype that calls a model endpoint with a parameterized prompt template

Implement chunking and embedding of a small document set and retrieve top-k passages for a query

Add basic input/output validation and an error-handling path to the prototype

Record latency and token usage for at least ten test queries

Extend the prototype to accept image input and describe the multimodal request/response flow

Compare speech-to-text and text-to-speech pipeline steps and where each can fail

Write five adversarial prompts and document how the system responds

Add a simple guardrail and re-run the adversarial set to compare results

Use an AI coding assistant to refactor one module of the prototype and review every suggested change

Explain how to evaluate AI-generated code for correctness, security, and maintainability

Describe model deployment options and how to choose between them for a given workload

Set up a minimal monitoring view for the prototype (error rate, latency, request volume)

Revisit the two domains you scored lowest on and rewrite your notes for those areas

Answer 30 mixed practice questions under timed conditions

For every missed question, identify the underlying concept gap and write a corrective note

Explain aloud, without notes, how a full generative AI application moves from data to deployed endpoint

Condense all notes into a single two-page review sheet organized by domain

Re-run the week 4 and week 5 evaluations to confirm the prototype still behaves as documented

Take one full-length timed practice exam and review every explanation

List remaining weak areas and schedule a final 90-minute review before exam day