04 - OCI AI services and infrastructure¶

The service selection questions are the fastest marks on the exam.

The four layers¶

Identifying the layer usually resolves the question. "Without machine learning expertise" points at a ready-made service; "train a custom model" points at Data Science; "distributed training across many GPUs" points at infrastructure.

Ready-made AI services¶

OCI Language - Sentiment analysis, at document and aspect level - Named entity recognition - Key phrase extraction - Language detection - Text classification - Translation - PII detection and redaction - Custom models for classification and entity recognition

OCI Speech - Speech to text transcription, batch and real time - Multiple languages, punctuation, and profanity filtering - Speaker diarization

OCI Vision - Image classification and object detection - Text detection in images (OCR) - Face detection - Custom models trained on your own labeled images

OCI Document Understanding - Extraction from forms, invoices, receipts, and identity documents - Table extraction, key-value extraction, document classification

Generative AI¶

OCI Generative AI provides managed access to foundation models through an API and console: - On-demand inference for variable workloads - Dedicated AI clusters for predictable throughput and isolation - Custom models through parameter-efficient fine-tuning on your own data - Embedding models for semantic search and RAG - Content moderation controls

OCI Generative AI Agents provides retrieval-augmented agents over enterprise data sources, so users can ask questions in natural language and get answers grounded in company documents, with citations.

AI Vector Search in Autonomous Database stores embeddings alongside relational data and performs similarity search in SQL, which removes the need for a separate vector store.

Select AI lets users query the database in natural language, with the model generating the SQL.

ML platform¶

OCI Data Science is the environment for building your own models: - Notebook sessions - managed JupyterLab with GPU options - Model catalog - versioned model storage with metadata and provenance - Model deployment - models served as HTTP endpoints with autoscaling - Jobs - repeatable training or processing runs - Pipelines - orchestrated multi-step ML workflows - AI Quick Actions - deploy, fine-tune, and evaluate foundation models without writing serving code - Feature store and model monitoring capabilities

OCI Data Labeling creates labeled datasets for supervised training, supporting images, text, and documents.

AI infrastructure¶

• GPU shapes on virtual machines and bare metal for training and inference
• Bare metal where you need the full device without virtualization overhead
• RDMA cluster networking for low-latency, high-throughput communication between nodes during distributed training. This is the differentiator for large-scale training, because the interconnect is the bottleneck once you exceed one node
• Storage: Object Storage for datasets, File Storage for shared access during training, Block Volumes for attached performance

Responsible AI in OCI¶

• Data used with OCI AI services is not used to train Oracle's base models
• Dedicated AI clusters provide isolation for sensitive workloads
• Content moderation controls in Generative AI
• PII detection in OCI Language supports privacy workflows
• Standard OCI controls apply: IAM policy, compartments, private endpoints, encryption, and audit logging

Service selection drill¶

Key terms¶

• OCI Language - the ready-made service for sentiment, entities, key phrases, translation, and PII detection
• OCI Speech - the ready-made service converting speech to text
• OCI Vision - the ready-made service for image classification, object detection, and OCR
• OCI Document Understanding - the service extracting structured data from forms and documents
• OCI Generative AI - the managed service providing foundation model inference, embeddings, and fine-tuning
• Dedicated AI cluster - isolated capacity for predictable generative AI throughput
• OCI Generative AI Agents - the managed service providing retrieval-augmented agents over enterprise data
• AI Vector Search - Autonomous Database's native vector storage and similarity search
• Select AI - the capability translating natural language questions into SQL against the database
• OCI Data Science - the platform for building, training, deploying, and monitoring custom models
• Model catalog - the versioned store for models with metadata and provenance in OCI Data Science
• Model deployment - serving a cataloged model as a scalable HTTP endpoint
• AI Quick Actions - OCI Data Science capability for deploying and fine-tuning foundation models without custom serving code
• OCI Data Labeling - the service for creating labeled datasets for supervised training
• Bare metal GPU shape - a compute shape providing direct access to GPUs without virtualization overhead

Related¶

• Notes 01: AI and ML fundamentals
• OCI Generative AI Professional
• Service comparison: AI and ML