Prepare and test your skills
Prepare and test your skills
Personalized generative AI solutions are systems that use conversational interfaces and intelligent search to give tailored, context-aware answers based on enterprise data. In Google Cloud, foundation models (opens in a new tab) are large models trained on vast amounts of text, images, video, and audio that predict likely responses and generate new content. Google offers foundation models through a managed API, and the Vertex AI model garden provides access to available models. Key building blocks for personalized solutions are embeddings, retrieval-augmented generation (RAG), and function calling.
Generative AI virtual agents are conversational systems that synthesize information from many enterprise sources into specific, reasoned, and actionable responses. A virtual agent can draw on historical support cases, websites, product documentation, inventory, emails, and old chat conversations that include resolutions. Virtual agents understand conversational queries, summarize complex information, and base answers on the provided data in line with business expectations.
Generative AI chatbots differ from traditional AI chatbots in understanding, inputs, and resolution. Traditional chatbots focus on specific keywords and are limited to keyword matching. Generative AI chatbots interpret complex language and full context and accept text, voice, and image input. For example, a customer can share a photo of a damaged product, and the chatbot combines the description with the photo to improve troubleshooting accuracy.
| Chatbot type | How it understands a request | Input methods | Effect on issue resolution |
|---|---|---|---|
| Traditional AI chatbot | Focuses on specific keywords | Text-based keyword input | Limited to keyword matching |
| Generative AI chatbot | Interprets complex language, nuances, and full context | Text, voice, and image | Combines all input to improve diagnostic accuracy |
Exam tip: A traditional AI chatbot matches keywords, while a generative AI chatbot interprets full context and can combine text, voice, and images.
Intelligent search for personalized needs is a meaning-based retrieval system that uses embeddings and RAG to find information by meaning rather than exact words. An embedding is a numerical representation of data such as text, images, or video that captures relationships between inputs. Embeddings are created during training by converting data into arrays of floating point numbers called vectors. Vectors map data into a continuous vector space where closer proximity means similar meanings, and reduced dimensionality improves efficiency for large datasets.
RAG systems use embeddings to match a user query with relevant enterprise knowledge. The RAG retrieval flow runs in order: first the user query is converted into an embedding, then the query embedding is compared with document embeddings in a knowledge base, then a similarity search retrieves the most semantically relevant documents for the response. This flow lets the system pull from enterprise knowledge instead of relying only on the model’s internal knowledge, which supports tailored and context-aware interaction.
A five-step process flow showing the RAG retrieval flow: a user query is converted into an embedding, compared with document embeddings in a knowledge base, a similarity search retrieves the most semantically relevant documents, the query and documents are sent to the model, and the model returns a context-aware response grounded in trusted enterprise data.
Function calling is the mechanism that connects large language models (LLMs) to external tools such as APIs and functions. Function calling lets an LLM use real-time information from databases, customer relationship management systems, and document repositories. It is a critical part of AI agents because it gives the model a structured way to choose a tool and format input for precise interaction.
A large context window lets a personalized solution handle large amounts of user or enterprise context. A context window of 1 million tokens can process any one of the following inputs:
The case management and insight-generation workflow is the ordered process that turns a chatbot conversation into a prioritized and routed support ticket. A generative AI-powered chatbot can automatically generate tickets from every interaction and analyze urgency, sentiment, and complexity to support prioritization. The chatbot also interfaces directly with the support ticketing system to create and populate the ticket with required information.
When the chatbot populates the ticket, it can include customer details, technical categorization and priority, a full conversation transcript, and a summary of the core issue. For new and complex issues, the supporting context includes a summary of the issue and conversation so the correct team starts with the right background.
Evaluating solution fit for personalized needs means matching the kind of user need to the generative AI capability that meets it. Conversational virtual agents fit support interactions that need context-aware dialogue and multimodal input. Intelligent search with embeddings and RAG fits needs where users must find answers synthesized from enterprise knowledge. Function calling fits needs where a response depends on real-time external data or actions in databases, CRM systems, or document repositories. The case management workflow fits needs where every interaction must become a prioritized ticket routed to the right team. These capabilities can work together when a conversational interface uses intelligent search for knowledge, function calling for external systems, and case management to track the interaction.
Generative AI modalities are the different kinds of output a model can produce, such as text, images, code, video, music, or numerical embeddings. Each modality maps to specific Google Cloud models and functions, and choosing the right one starts with the output the task requires. Google Cloud provides Gemini models for text and multimodal work, Imagen for images, Veo for video, Lyria for music, and embedding models for search and comparison.
Text generation is the production of written content by passing a prompt to a large language model (LLM). In BigQuery, AI.GENERATE (opens in a new tab) is the most flexible inference function and can analyze text, image, audio, video, or PDF data, while AI.GENERATE_TEXT is a table-valued version that also supports partner models such as Anthropic Claude, Llama, and Mistral AI, and open models. Both functions share the same purpose, billing model, and scalability and process between 1 million and 10 million rows per 6-hour query job. AI.GENERATE is a scalar function that returns a single value and needs no remote model, while AI.GENERATE_TEXT takes a table, returns a table, and requires a remote model to be created first.
Managed AI functions handle routine text tasks with streamlined syntax and no customization. AI.IF filters data with natural language conditions, AI.SCORE rates inputs to rank rows by quality or similarity, AI.CLASSIFY sorts inputs into user-defined categories, and AI.AGG summarizes or analyzes data. Each managed function returns a single scalar value such as a BOOL, FLOAT64, or STRING. Task-specific functions instead call Cloud AI APIs for natural language processing, machine translation, document processing, audio transcription, and computer vision.
Image generation is the creation or editing of visual assets, and Imagen is the Google Cloud model for this work. Imagen can edit image content using a mask, add content with mask-based inpainting, add content with automatic mask detection without a manual mask, and produce short-form image captions. Image analysis is a separate task that describes images that already exist. Cloud Vision API detects text in images using the TEXT_DETECTION or DOCUMENT_TEXT_DETECTION enums, and AutoML Vision object detection converts sections of images into text documents.
Exam tip: Imagen generates and edits new images, while Vision AI and Cloud Vision API analyze images that already exist.
Code generation is technical development output such as code predictions and unit tests. Pre-trained code generation models support batch code prediction for development tasks. Text embedding models can calculate embeddings for code blocks and queries to support code retrieval tasks. Prompts can also be created to work with a publisher chat model to generate unit tests.
Video, music, and embeddings are three different generative AI outputs for media creation and for search and comparison. Veo generates high-resolution videos in a wide range of cinematic and visual styles. Lyria generates high-quality music with instrumentals and vocals, including transformation, continuation, and control of style and performance. An embedding is a high-dimensional numerical vector that represents text, image, audio, video, or document data.
AI.EMBED creates an embedding from text or image data, and AI.GENERATE_EMBEDDING adds a column of embedded text, image, audio, video, or document data to a table. Embeddings support RAG, which augments model responses with data from a trusted source for better factual accuracy and access to data newer than training data. Embeddings also enable multimodal search with text input to search images, semantic search for recommendations and record deduplication, and k-means clustering.
Matching modalities to models means selecting the Google Cloud model or function based on the output type the task produces. The choice depends on whether the business needs written content, visual assets, code, media, or vectors for search.
| Modality | Google Cloud model or function | Output produced |
|---|---|---|
| Text | Gemini, AI.GENERATE, AI.GENERATE_TEXT | Written content, answers, summaries, structured data |
| Image | Imagen | Generated or edited images, image captions |
| Code | Pre-trained code generation models | Code predictions, unit tests, code retrieval embeddings |
| Video and music | Veo, Lyria | Videos, music with instrumentals and vocals |
| Embeddings | AI.EMBED, AI.GENERATE_EMBEDDING | Numerical vectors for search, comparison, and clustering |
A single task can combine modalities when a text prompt is embedded and compared against image embeddings for multimodal search. RAG can also feed trusted text into a model to personalize responses for a user query.
Aligning AI solution types with business use cases means matching what a model produces to a business goal such as operational efficiency, creative innovation, or enhanced customer engagement. BigQuery AI functions (opens in a new tab), Vertex AI models such as Gemini and Imagen, Document AI, Cloud Vision API, Natural Language API, and Cloud Translation API are the main building blocks. The choice depends on the output the business needs and whether the task is general or narrow.
Text generation for content and analysis uses large language models (LLMs) to create, summarize, classify, or extract information from text and unstructured data. In BigQuery, AI.GENERATE is a flexible scalar function that sends a prompt to a Gemini model and returns generated text, a full model response in JSON, and call status. AI.GENERATE_TEXT is a table-valued function that takes a table as input and returns a table as output, and it supports Gemini models, partner models such as Anthropic Claude, Llama, and Mistral AI, and open models. Both can process between 1 million and 10 million rows per 6-hour query job, depending on token length.
AI.GENERATE_TEXT depends on a remote model and a BigQuery connection with the Agent Platform User role, while AI.GENERATE can be called directly with end-user credentials. For structured output, a SQL schema is specified with AI.GENERATE or AI.GENERATE_TABLE, while AI.GENERATE_BOOL, AI.GENERATE_DOUBLE, or AI.GENERATE_INT produce a single output field. Managed functions simplify routine tasks where BigQuery selects the model to optimize cost and quality:
Text generation fits content creation, question answering, sentiment analysis, summarization, and structured extraction. General-purpose functions fit needs for control over model, prompt, and parameters, while managed functions fit routine filtering, rating, classification, or aggregation. Optimized mode for AI.IF and AI.CLASSIFY trains a lightweight distilled model on a sample of data, then uses that model for inference on most rows to reduce token use and latency for thousands or billions of rows.
Image generation and visual understanding separate creating new visuals from analyzing existing images. Imagen on Vertex AI generates short image captions and edits images using masks, where mask-based editing targets a specific area and automatic mask detection lets Imagen find the area. Gemini models can describe an image or analyze image data. Cloud Vision API converts image content to text using optical character recognition (OCR) and detects handwriting, objects, image properties, and safe search. Document AI Enterprise Document OCR also converts images to text.
Business alignment depends on the desired output and goal. Creative innovation uses Imagen to create marketing visuals or edit product images. Customer engagement uses captions, visual search, or image descriptions. Operational efficiency uses OCR and object detection to turn scanned documents or images into usable data. Choose Imagen when the output is a new or edited image, and choose Gemini, Cloud Vision, or Document AI when the output is text, labels, or extracted data.
Code generation and developer productivity cover specialized text generation where the model helps write, test, or retrieve code. Vertex AI samples show prompts that generate unit tests with a publisher chat model. Text embedding models calculate embeddings for code blocks and queries for code retrieval tasks. Function calling can let models query external tools, such as the Google Places API, to enhance responses. Business alignment focuses on operational efficiency when developers generate tests, search code semantically, or automate repetitive coding tasks. Creative innovation can also use code generation to prototype new features faster. Choose code generation when the goal is software delivery speed or developer assistance, not customer-facing content or image creation.
Personalization and customer engagement tailor responses and recommendations to user needs with embeddings, vector search, and RAG. An embedding is a high-dimensional numerical vector that represents text, images, audio, video, or documents. AI.EMBED creates an embedding from text or image data, and AI.GENERATE_EMBEDDING adds a column of embedded data to a table. Embeddings support RAG to augment responses with trusted data, multimodal search such as text input to search images, semantic search for recommendations and deduplication, and clustering with a k-means model.
RAG augments model responses by referencing additional trusted data for better accuracy. The RAG flow runs in order: first a user query arrives, then relevant trusted data is retrieved through embeddings and vector search, then the query and retrieved data are sent to the model, then the model returns a more factually accurate response that can use data newer than training data. Enhanced customer engagement uses personalized recommendations, semantic search, and ticket classification, while operational efficiency uses AI.CLASSIFY to route tickets by billing, shipping, or product quality and AI.SCORE to rank reviews by sentiment or quality.
Task-specific AI for operational efficiency uses Cloud AI APIs to perform narrow, pre-trained tasks without building a general-purpose model. Natural Language API classifies content with a generalized list of categories and identifies common public entities. AutoML Natural Language creates custom models for classification or entity extraction with company-specific labels. Cloud Translation API translates text between thousands of language pairs and can translate documents such as PDF and DOCX. Document AI Form Parser and pretrained processors extract known and patterned entities like phone numbers and addresses. Cloud Vision API handles OCR, handwriting, object detection, image properties, and safe search.
Task-specific AI fits automation of document processing, translation, transcription, and image analysis. For document classification, text-based methods such as Natural Language API or AutoML Natural Language are generally recommended over AutoML Vision classification, which tends to be less accurate for documents. Choose task-specific functions when the task is well-defined and a pretrained API is sufficient, and choose general-purpose or managed functions when the task needs flexible prompts, custom schemas, or model choice.
Choosing the right solution type means matching the business goal to the model output and the level of control required. Operational efficiency favors task-specific and managed functions, creative innovation favors image and text generation, and enhanced customer engagement favors embeddings, RAG, and personalization.
| Solution type | Primary output | Google Cloud capability | Business goal |
|---|---|---|---|
| Text generation | Text, summaries, classifications, structured data | AI.GENERATE, AI.GENERATE_TEXT, managed AI functions | Content creation, analysis, operational efficiency |
| Image generation and understanding | New or edited images; captions, labels, OCR text | Imagen, Gemini, Cloud Vision API, Document AI | Creative innovation, customer engagement, document automation |
| Code generation | Code, unit tests, code embeddings | Vertex AI publisher chat models, text embedding models | Developer productivity, faster software delivery |
| Personalization | Embeddings, recommendations, ranked or classified results | AI.EMBED, AI.GENERATE_EMBEDDING, AI.SCORE, AI.CLASSIFY, RAG | Enhanced customer engagement, semantic search |
| Task-specific AI | Translation, entities, transcription, vision labels | Natural Language API, Cloud Translation API, Document AI, Cloud Vision API | Operational efficiency for narrow, repeatable tasks |
Exam tip: Use general-purpose AI functions when you need control over the model, prompt, and parameters; use managed AI functions for routine filtering, rating, classification, or aggregation where BigQuery selects the model to optimize cost and quality.
A traditional AI chatbot focuses on specific keywords and is limited to keyword matching with text-based keyword input, while a generative AI chatbot interprets complex language, nuances and full context and accepts text, voice and image input. For example, a generative AI chatbot can combine a customer description with a photo of a damaged product to improve troubleshooting accuracy.
Use Imagen on Vertex AI when you need to generate or edit images, including mask-based editing of a specific area, automatic mask detection without a manual mask, and short-form image captions. Use Cloud Vision API or Document AI when you need to analyze existing images to produce text, labels or extracted data, such as optical character recognition (OCR), handwriting, object detection, image properties and safe search.
AI.GENERATE in BigQuery is a scalar function that returns a single value and needs no remote model, while AI.GENERATE_TEXT is a table-valued function that takes a table as input, returns a table as output and requires a remote model to be created first. Both share the same purpose, billing model and scalability and process between 1 million and 10 million rows per 6-hour query job. AI.GENERATE_TEXT also supports partner models such as Anthropic Claude, Llama and Mistral AI, and open models.
Function calling connects large language models (LLMs) to external tools such as APIs and functions by letting the model generate structured data with a function name and parameter values that calls the function. The function returns its results to the model, and the model incorporates the results into its reasoning to generate a response using real-time information from databases, customer relationship management systems and document repositories.
An enterprise strategy department spends considerable time reviewing extensive market analysis documents and manually transforming the core insights into structured presentation slide outlines and executive summaries for leadership. The organization aims to enhance operational efficiency by automating these content drafting tasks so employees can focus on strategic decision-making.
Which generative AI solution type should the organization implement to achieve this goal?
Last updated