Skip to content

Oracle Cloud Infrastructure Generative AI Professional Fact Sheet

Exam Overview

Exam Code: 1Z0-1127-25 Exam Name: Oracle Cloud Infrastructure Generative AI Professional Level: Professional Duration: 90 minutes Format: Multiple choice Questions: 60 Passing Score: 68% Cost: USD 245 list price; Oracle periodically runs free certification windows Valid For: Re-versioned annually, indicated by the year suffix in the exam code Delivery: Oracle CertView, online proctored Prerequisites: None formally; OCI AI Foundations and Python familiarity strongly recommended

Verify before booking. Oracle re-versions this exam annually and the OCI Generative AI service changes quickly. Confirm the current exam code and objectives before building a study plan around this sheet.

📖 OCI Generative AI Professional - exam page and objectives 📖 OCI Generative AI documentation - service reference 📖 Oracle University free learning path - the official course, usually free

What this exam covers

Three areas, roughly equal in weight:

  1. Fundamentals of large language models - architecture, prompting, fine-tuning, decoding
  2. Using the OCI Generative AI service - models, inference, dedicated clusters, custom models, security
  3. Building an LLM application - RAG, LangChain, vector stores, deployment, and evaluation

It is a builder's exam. Where OCI AI Foundations asks what things are, this asks how you would assemble them, and expects familiarity with code-level concepts even though there is no coding on the exam itself.

Target Audience

  • Developers building generative AI applications on OCI
  • Solution architects designing LLM systems
  • Data scientists moving into LLM application work
  • Anyone who has passed OCI AI Foundations and wants depth

Exam Domains

Fundamentals of large language models

Key Concepts: - Transformer architecture, attention, encoder and decoder variants - Tokenization and the relationship between tokens, cost, and context limits - Embeddings, embedding models, and semantic similarity - Decoding parameters: temperature, top-k, top-p (nucleus), frequency and presence penalties, max tokens, stop sequences - Greedy decoding versus sampling, and the effect on determinism - Prompt engineering: zero-shot, few-shot, chain-of-thought, system prompts - Prompt injection and mitigation at a conceptual level - Hallucination, grounding, and citation - Model customization spectrum: prompting, RAG, parameter-efficient fine-tuning (T-Few, LoRA), full fine-tuning, pre-training - Evaluation: loss, perplexity, accuracy on held-out sets, and LLM-as-judge

Using the OCI Generative AI service

Key Concepts: - Available foundation models: chat, embedding, and their context limits - On-demand inference versus dedicated AI clusters, and when each is appropriate - Cluster unit sizing, hosting versus fine-tuning clusters - Custom model creation through fine-tuning, and the endpoint that serves it - The playground, API, SDK, and CLI surfaces - Security: compartments, IAM policy, private endpoints, encryption, customer data isolation - Content moderation controls - Monitoring, limits, and quotas - OCI Generative AI Agents for retrieval-augmented agents over enterprise data

Building, deploying, and evaluating LLM applications

Key Concepts: - RAG architecture end to end: ingest, chunk, embed, store, retrieve, augment, generate - Chunking strategy and its effect on retrieval quality - Vector stores, including Oracle AI Vector Search in Autonomous Database - Similarity metrics: cosine, dot product, Euclidean - LangChain concepts: models, prompts, chains, memory, retrievers, document loaders, output parsers - Conversation memory and its effect on context and cost - Chatbot design patterns - Evaluation of RAG: retrieval quality and answer quality - Observability, tracing, and cost management - Deployment on OCI and integration with other OCI services

The customization decision

The most testable decision on the exam.

Approach Changes Cost Choose when
Prompt engineering Nothing about the model Lowest Always start here
Few-shot prompting Nothing; supplies examples in context Low The task needs demonstration, not new knowledge
RAG What the model knows for this request Low to medium The model needs access to your data, which changes
Fine-tuning (T-Few / LoRA) Model behavior, style, format adherence Medium Consistent tone or output format is required, and prompting is not enough
Full fine-tuning All model weights High Rare; a large, stable, domain-specific dataset
Pre-training Builds a model from scratch Very high Effectively never, outside model providers

Rule the exam applies: knowledge problems are RAG problems; behavior problems are fine-tuning problems.