Oracle Cloud Infrastructure 2025 Generative AI Professional (1Z0-1127-25) Study Guide

Domain 1: Fundamentals of Large Language Models

Key concepts you must know · 168 practice questions

LLM architectures: decoder-only (GPT-style, autoregressive), encoder-only, and encoder-decoder models.
Transformer fundamentals: self-attention, multi-head attention, positional encoding, and context windows.
Tokenization, tokens versus words, and how token limits affect cost and context.
Decoding and sampling: greedy decoding, beam search, and the effect of temperature, top-k, and top-p on output.
Prompt engineering: zero-shot, few-shot in-context learning, chain-of-thought, and system versus user prompts.
Fine-tuning fundamentals (PEFT) and when to fine-tune versus prompt versus use RAG, plus responsible-AI basics.

Domain 2: Using OCI Generative AI Service

Key concepts you must know · 216 practice questions

The fully managed OCI Generative AI service and its pretrained chat models (Cohere Command R, R+, A; Meta Llama 3.x) and embedding models (Cohere Embed).
The Playground for testing prompts, and inference parameters (temperature, top-p, top-k, max tokens, penalties).
Dedicated AI Clusters: fine-tuning clusters (train custom models) versus hosting clusters (serve endpoints), single-tenant and scaled by units.
Fine-tuning methods: T-Few (parameter-efficient, smaller datasets) versus Vanilla, with a JSONL dataset of at least 32 pairs split 80/20.
Creating model endpoints to serve pretrained and custom models for inference.
Security: IAM policies, private endpoints, data isolation in dedicated clusters, and choosing on-demand versus dedicated capacity.

Domain 3: Implement RAG using OCI Generative AI Service

Key concepts you must know · 174 practice questions

The retrieval-augmented generation workflow: retrieve, augment, and generate, and why it grounds responses.
Document ingestion: loading, splitting, and chunking documents with size and overlap tradeoffs.
Creating embeddings of chunks with the OCI Generative AI embedding model.
Storing and indexing vectors in Oracle Database 23ai with AI Vector Search, and similarity search to retrieve top-k chunks.
Generating a grounded answer by feeding retrieved context plus the query to the chat model.
Building RAG and a conversational chatbot with LangChain: prompt templates, chains, memory, loaders, splitters, and vector stores.

Domain 4: Using OCI Generative AI RAG Agents Service

Key concepts you must know · 144 practice questions

The fully managed OCI Generative AI Agents service and what a RAG agent is (retrieve for accuracy, generate for fluency).
Creating knowledge bases from OCI Object Storage or bring-your-own vector search (Oracle Database 23ai, MySQL HeatWave, OCI OpenSearch).
Creating, deploying, and invoking an agent as a multi-turn chatbot with source attribution.
Agent tools: the RAG Tool (hybrid search), the SQL Tool (natural language to SQL), the Agent Tool (multi-agent), and custom function and API tools.
Guardrails: content moderation, prompt-injection protection, and PII protection at agent endpoints.
Custom instructions and optional human-in-the-loop monitoring.

Oracle Cloud Infrastructure 2025 Generative AI Professional (1Z0-1127-25) exam tips

Using OCI Generative AI Service is the heaviest domain. Know the pretrained model families (Cohere Command and Meta Llama), the difference between fine-tuning and hosting dedicated clusters, and T-Few versus Vanilla fine-tuning, including the 32-pair JSONL and 80/20 split.
Master the decision between fine-tuning, RAG, and prompt engineering - the exam repeatedly asks which fits a scenario. RAG grounds answers in your data without retraining; fine-tuning adapts the model; prompting is the lightest option.
For RAG, know the pipeline end to end: chunk, embed, store and index vectors in Oracle Database 23ai, retrieve by similarity, then generate. This exam uses Oracle Database 23ai (not newer branding) and the Cohere and Llama 3.x model families.
For the Agents service, distinguish the tools: the RAG Tool retrieves from a knowledge base with hybrid search and source attribution, the SQL Tool converts natural language to SQL, and the Agent Tool orchestrates multiple agents. Know the knowledge-base source options.
Understand decoding parameters precisely: higher temperature and top-p make output more diverse and less deterministic; greedy and low temperature make it more predictable. These are common single-fact questions.

Study guide FAQ

How many questions are on the exam and what is the passing score?

There are about 50 scored, scenario-based multiple-choice questions in 90 minutes, with a passing score around 68 percent. The exam is delivered through Pearson VUE online or at a test center, costs 245 USD, and the certification is valid for two years.

Do I need programming experience?

There is no formal prerequisite, but Oracle recommends familiarity with Python and basic machine-learning and deep-learning concepts. The exam is scenario-based around building LLM and RAG applications on OCI, so comfort with code helps even though you do not write large programs on the exam.

Which OCI services and tools are covered?

The OCI Generative AI service (pretrained Cohere Command and Meta Llama models, dedicated AI clusters, fine-tuning), retrieval-augmented generation with Oracle Database 23ai vector search, LangChain, and the OCI Generative AI Agents service with its RAG, SQL, and agent tools.

What changed in the 2025 version?

Oracle updated the exam to cover new service features and simplified it based on feedback. The most notable addition is a full domain on the OCI Generative AI RAG Agents service, which became generally available after the 2024 exam was authored.

Is CertGrid practice official Oracle material?

No. CertGrid is an independent practice platform and is not affiliated with or endorsed by Oracle. These questions are original and written to mirror the current exam objectives so you can rehearse them. Always confirm the current exam topics on Oracle's certification site before your exam.

Related Oracle Cloud resources

Test yourself: 702 Oracle Cloud Infrastructure 2025 Generative AI Professional (1Z0-1127-25) practice questions