Prepare and test your skills

Prepare and test your skills

The range of Google Cloud AI and ML solutions includes pre-trained APIs, Vertex AI (opens in a new tab) platform services, and custom model development tools. The three groups differ in who trains the model and how much work the user does. Pre-trained APIs need no training, Vertex AI manages the full model lifecycle, and custom tools learn from organization-specific data.
Core categories of Google Cloud AI solutions separate ready-made intelligence from models that learn from customer data. Pre-trained APIs are Google-managed services where data flows from the application to the API and back as a finished result. The application depends on the API for inference but does not manage training data or infrastructure. Vertex AI connects pre-trained capabilities and custom development in one managed environment where deployed models receive inference requests and return outputs to applications.
Common pre-trained APIs include:
Custom model development tools train on customer labels when pre-trained models are not accurate enough. AutoML Vision handles image classification and object detection, AutoML Natural Language handles custom classification and entity extraction, and BigQuery ML trains and runs models with GoogleSQL inside the warehouse. Deep Learning VM Images and Deep Learning Containers provide preconfigured environments for deeper custom work. The tradeoff is clear: pre-trained APIs need no training data or expertise but cannot learn company-specific labels, while custom tools need labeled examples and evaluation but recognize specialized patterns.
Pre-built services versus custom workflows is a choice between calling an API and building a unique model. Pre-built services accept a standard data type such as image, audio, text, video, or document and return a result without training. Data flows from the application to the service API and back, so the application depends on the service but avoids managing training. Custom workflows move data from storage into a development environment, run training on accelerated compute such as Cloud TPU, and deploy the resulting model for use.
A decision tree branching on API fit, training data availability, and control needs to choose pre-built APIs, AutoML or BigQuery ML, or fully custom training.
Custom environments include TensorFlow Enterprise for building TensorFlow models, Colab Enterprise for collaborative notebooks, Deep Learning VM for preconfigured virtual machines, and Deep Learning Containers for consistent containers. Agent platforms add a third pattern for connected outcomes where Gemini Enterprise Agent Platform builds autonomous agents, video storage feeds agents and analytics, and a knowledge graph consolidates organizational knowledge. Choose pre-built services when expertise is limited and speed matters, and choose custom workflows when no API matches the problem or large volumes of specialized data justify extra effort.
| Option | Data and customization | Expertise and cost |
|---|---|---|
| Pre-built AI services | Standard data types and small to moderate volumes; limited customization | Application skills, not training skills; lower upfront effort and faster value |
| Custom ML workflows | Large volumes of relevant training data; high customization | Machine learning expertise; higher effort for setup, training, and tuning |
Mapping business use cases to products means matching data type, customization need, and desired outcome. Document AI handles end-to-end document understanding, specialized APIs handle single tasks like vision or language, Vertex AI unifies custom training and inference, and BigQuery ML brings prediction directly to warehouse data. Vertex AI capabilities include batch inference that processes a group of requests into one output file, online inference that returns real-time results, Imagen for text-to-image generation and editing, and Vertex AI Search for finding relevant documents in a corpus.
Document processing converts dark data trapped in emails, PDFs, scanned forms, and receipts into usable fields. The pipeline runs in fixed order:
A three-step process flow showing documents flowing from sources such as Drive, scanners, and Cloud Storage into Document AI optical character recognition, then to classification and entity extraction, and finally to storage or downstream systems as usable fields. OCR must run first and classification runs second on the resulting plain text.
Optical character recognition must run first and classification runs second on the resulting plain text. The Cloud Natural Language API classifies with general categories and needs no training, while AutoML Natural Language trains on business labels. For prediction, BigQuery ML trains, evaluates, and runs inference inside the warehouse for forecasting, anomaly detection, classification, regression, clustering, and recommendations without moving data.
Exam tip: Speech-to-Text converts audio to text, while Text-to-Speech converts text to audio — the direction of the flow determines the service.
Decisions and tradeoffs for Google Cloud AI and ML selection balance speed, effort, differentiation, and required expertise. Fully managed services and pre-trained models shorten deployment timelines, while self-managed infrastructure and custom training give deeper control but take longer. Organizations choose well by matching the level of management to how fast they need results and what skills they can operate over time.
Speed and time-to-value tradeoffs compare how quickly a team moves from an idea to a working model in production. Vertex AI (opens in a new tab) is a fully managed, end-to-end platform for MLOps that offloads infrastructure management so teams focus on models and reach value faster. Cloud Run is a serverless option for containerized inference that scales to zero, which fits event-driven applications and smaller models without paying for idle capacity. Google Kubernetes Engine (GKE) is a self-managed Kubernetes option that gives deep control, portability, custom pipelines, and kernel-level customization, but that control requires more setup and operational effort.
Pre-trained models and APIs are the fastest path because they skip training from scratch. Training complex models from scratch needs significant compute and time, while starting from an existing model removes that step. Model Garden offers pre-trained models grouped as Google models such as Gemini and Imagen, open models such as Gemma and Llama, and third-party models from partners such as Anthropic. APIs such as Vision, Natural Language, Speech-to-Text, Video Intelligence, Document AI, Dialogflow, and Gemini let developers add capabilities in a fraction of the time of building a model. Fine-tuning sits in the middle because learning transfer applies knowledge from a pre-trained model to a new task with less data and compute.
| Service | Level of management | Best for |
|---|---|---|
| Vertex AI | Fully managed, end-to-end | Teams focused on MLOps and rapid time-to-value |
| Cloud Run | Serverless, scales to zero | Event-driven applications and smaller models |
| GKE | Self-managed Kubernetes | Deep control, portability, and custom pipelines |
Effort and operational complexity describe how much configuration, data preparation, and lifecycle maintenance each option requires. More control usually means more operational work, so the choice between Vertex AI, GKE, and Cloud Run directly affects feasibility. A team with limited Kubernetes expertise may find GKE harder to adopt, while a team that needs kernel-level changes may find Vertex AI or Cloud Run too restrictive.
GKE is chosen when the organization needs infrastructure control, large-scale training with minimal-latency serving, cost optimization with Spot VMs and Flex-start VMs, or portability with open standards. GKE reduces effort with GKE Autopilot (opens in a new tab) for automated operations and right-sizing, Inference Gateway for model-aware routing, faster startup nodes, Kueue for batch resource allocation, and TPU multislice for large-scale training. Data and lifecycle work adds effort everywhere through governance, Feature Store for reusable ML features, Model Registry for managing trained models, and pipelines with Agent Platform Pipelines, Kubeflow Pipelines, Cloud Logging, and Cloud Monitoring. Even with helpers, GKE still needs Kubernetes expertise and cluster maintenance, while Vertex AI offloads infrastructure and Cloud Run removes server management.
Exam tip: Choose Vertex AI or Cloud Run when rapid time-to-value matters most, and choose GKE when deep control, portability, or custom pipelines matter more than speed.
Differentiation and licensing expertise compare standardized capabilities with reuse of existing investments. On-Demand Licenses from Google Cloud provide standardized capabilities with low operational effort, while Bring Your Own License (BYOL) lets an organization apply existing Microsoft licenses on Google Cloud for potential cost and hybrid-cloud differentiation but requires deeper licensing expertise. On-demand licenses are bought directly from Google Cloud from prebuilt images, and the hourly fee includes both Compute Engine infrastructure and the Microsoft license for products such as Windows Server and SQL Server.
BYOL differentiation comes from cost control and deployment flexibility rather than a prebuilt image. When a BYOL-eligible license is used, the organization is billed only for Compute Engine infrastructure, with no additional license cost from Google Cloud. Savings are possible with active Software Assurance, especially for long-term stable workloads, and License Mobility helps move application servers such as SQL Server to Google Cloud. Licenses that need dedicated hardware with sole-tenant nodes can add infrastructure cost, and the organization remains responsible for Microsoft compliance. On-demand needs mainly basic cloud deployment skills because Google handles compliance and updates, while BYOL needs licensing and legal skills to validate eligibility and hardware requirements.
Pre-trained APIs, AutoML, and custom models are three paths for adding AI to different business use cases. Pre-trained APIs add common capabilities without training, AutoML builds custom models from proprietary data with less expertise, and fully custom development gives total control over code and infrastructure. The best choice depends on data volume, need for control, and performance goals.
Capabilities of pre-trained APIs let developers add machine learning through a simple request and response without collecting training data or managing training compute. A developer sends input such as an image, text, audio, or document to the API and receives structured results back. Because training is already done on vast datasets, pre-trained APIs reduce cost and development time and let a team move quickly from idea to working feature. A single application can combine several APIs, such as sending a document to one API for extraction and then sending the extracted text to another API for analysis.
| API | What it does | Typical use cases |
|---|---|---|
| Cloud Vision API | Derives insights from images | Image analysis, content moderation, and automated data entry |
| Cloud Natural Language API | Analyzes text structure and meaning | Customer feedback analysis and content categorization |
| Speech-to-Text API | Converts audio to text | Transcription across languages and dialects |
| Video Intelligence API | Identifies objects, scenes, and actions in video | Video analysis, moderation, and search |
| Document AI API | Extracts and classifies data in documents | Automating document workflows |
| Dialogflow API | Builds chatbots and voice assistants | Customer service bots |
| Gemini API | Provides general-purpose AI | General AI tasks |
The Cloud Natural Language API offers five analysis types: sentiment analysis for emotional tone, entity analysis for people and places mentioned, entity sentiment analysis for sentiment per entity, content classification for sorting text into categories, and syntax analysis for grammatical structure. Choose a pre-trained API when a common task must ship quickly without machine learning expertise, and choose AutoML or custom paths when the model must learn company-specific labels.
Custom model development means training and running a model with full control over data, code, and infrastructure. Custom development packages training code into custom containers stored in Artifact Registry for consistent and reproducible training, and trained models are stored in Model Registry for reuse in later pipelines or online prediction. Training complex models from scratch needs significant time and compute, so teams start with pre-trained models from Model Garden and use fine-tuning to apply knowledge to a new task before committing to full custom builds.
The platform choice balances management and control. Vertex AI offloads infrastructure for rapid development and MLOps, GKE keeps deep control and portability for very large training and minimal-latency serving, and Cloud Run removes server management for containerized inference that scales to zero. Cloud Run hosts AI applications by deploying code or a container as a scalable API endpoint where the application sends data to a model such as Gemini, Llama, Gemma, or a custom model and returns results. GKE supports GPUs and TPUs for inference, Inference Gateway for routing, and TPU multislice where multiple slices communicate over the data center network for large-scale training. Finding the efficient configuration follows an ordered process:
Exam tip: Fine-tuning a pre-trained model needs less data and compute than training from scratch, so start with pre-trained models and choose custom training only when control or performance demands it.
The role of AutoML in custom solutions is to build high-quality models from proprietary data without a large data science team. AutoML is the middle path between a ready-made API and fully custom code in Python or Java. BigQuery ML (opens in a new tab) is an example of this approach because SQL practitioners can create and run models directly where data already lives. A BigQuery ML model represents what the system learned from training data, and models are stored in BigQuery datasets like tables and views.
Bringing ML to the data speeds time to market because large datasets do not move to a separate environment. BigQuery ML increases development speed by avoiding data movement and formatting for Python frameworks, reduces complexity with fewer tools, and supports forecasting, anomaly detection, classification, regression, clustering, and recommendations. Teams work through the console, query editor, bq command-line tool, REST API, or Colab Enterprise notebooks, using CREATE MODEL with default settings and inference functions. Basic knowledge of feature engineering still helps optimize results, and Feature Store centralizes features for reuse across projects. Choose the custom-model path when predictions must learn from proprietary datasets such as sales history or customer behavior, and choose Cloud AI APIs when a ready-made task such as translation fits.
CREATE MODEL.Pre-trained APIs require no training by the user, Vertex AI provides a managed platform for building, training, deploying and running inference, and custom tools train models on organization-specific labels. Pre-trained APIs receive application data and return a finished result without managing training data or infrastructure, while custom workflows move data into a development environment, train on accelerated compute such as Cloud TPU, and deploy the resulting model for use.
Use pre-built AI services in Google Cloud when expertise is limited and speed matters, and use custom ML workflows when no API matches the problem or large volumes of specialized data justify extra effort. Pre-built services handle standard data types such as image, audio, text, video or document with lower upfront effort and faster value, while custom workflows need large volumes of relevant training data and machine learning expertise with higher effort for setup, training and tuning.
Optical character recognition must run first and classification runs second on the resulting plain text in Google Cloud document processing. Documents flow from sources such as Drive, scanners or Cloud Storage to optical character recognition, recognized text flows to classification or entity extraction, and extracted fields flow to storage or downstream systems for validation and analysis.
Use BigQuery ML in Google Cloud when predictions must learn from proprietary datasets such as sales history or customer behavior without moving data. BigQuery ML trains, evaluates and runs inference with GoogleSQL inside the warehouse for forecasting, anomaly detection, classification, regression, clustering and recommendations, letting SQL practitioners build custom models where data already lives.
Last updated