OS AI Agentic Practice Test

Which of the following options is not a Characteristic of an AI Agent?. The ability to function as an orchestrator by leveraging the LLM's reasoning to autonomously manage a complex workflow needed to complete a user request. The agent's core distinction is its role as a passive text processor, solely dedicated to optimizing the summarization and translation of documents based on a user's prompt. The necessity of agency, which includes breaking down objetives, forming multi-steps plans, utilizing external tools and integrating feedback. ] Operating as a proactive problem-solver designed not just to provide answer, but also to execute actions for real-world tasks.

What do the Reasoning and Decision capabilities bring to an AI Agent?. The tracking of calls, tokens, response quality, latency and overall agent health. The ability to interpret context and make informed decisions. The controls to enforce safe,compliant and const-effective AI operations includig usage limits and access permissions. The mechanisms that enable agents to retain context and learn from past interactions, thereby improving their performance over time.

Which prompting technique involves asking an LLM to perform a task without providing any prior examples?. Self-Consistency Prompting. Zero-Shot Prompting. Chain-of-Thought Prompting. Hallucinations.

Which major limitation of LLMs involves generating plausable but factually incorrect or nonsensical information?. Reasoning Errors. Knowledge Cutoff. Bias. Hallucinations.

Considering the default service action inputs (UserInput and SessionId) in a typical Agent App. Which of the following sentences is true?. The UserInput is used in the System Prompt to define the Agent's behaviour. The SessionId is used to filter the Memory data. The UserInput overrides the System Prompt. The SessionId is used to filter the AI model's response.

What characterizes a Sequential Agent Pattern in a Multi-Agent System?. Agents operate in a reflection loop, iteratively reviewing and improving a single, shared document output. Multiple agents are triggered simultaneously and run after independent tasks to obtain diverse outputs. A complex task is broken down into steps, and the output from one specialized agent becomes the direct input for the next agent. A central Orchestration Agent manages the entire workflow by delegating all sub-tasks to specialized worker agents.

What are the basic units of texts (words, sub-words or characters) in the context of LLMs?. Embeddings. Tokens. Parameters. Weights.

When an Agent requires persistent data outside its conversational scope or specific preferences across multiple sessions, which memory type, often implemented using a Vector Database, should be designed into the Agent's architecture?. Knowledge Base Memory (RAG). Short-Term Memory (Conversational). Long-Term Memory (Persistent). Working Memory (Scratchpad).

In cases where an Agent App needs to run its logic asynchronously due to a long-running model call (e.g. generating too many tokens or using thinking models) that might otherwise exceed normal timeouts, which element can be used to run the Agent with an extended timeout of 5 minutes?. Events. An External System call. A Client Action. A Timer.

What is the recommended approach to integrate a proprietary AI Search Service (e.g. a specific vector database) into the ODC ecosystem for use in a RAG-based Agentic Application?. Implement the integration using a regular app, and expose via Service Actions. Then consume the Service Actions directly in your Agentic app. Implement the integration using a regular app, and expose it using a Rest API. Then use that endpoint to set up a Custom Ai Integration in the ODC Portal. Implement the integration using a library and expose it using Server Actions. Then consume the Server Actions directly in your Agentic app. By configuring the connection of the AI Search Service directly within the Agentic Settings.

In the high-level RAG workflow, which operation is performed after the relevant context is retrieved from the Knowledge base?. The context is added to prompt and sent to the LLM. A hybrid search is performed against the database. The user prompt is converted into vectors. The LLM updates the external Knowledge Base.

In the Agent Parallel Processing architecture pattern, how multiple simultaneous calls to the same agent executed to reduce overall latency?. By ensuring the agent is deployed as a Single Monolithic Agent. By leveraging Events to launch a new separate thread for each concurrent call. By implementing a Queue mechanism that serializes requests. By utilizing the Workflow built-in Parallel flow element.

What is the main reason Text Chunking is considered an important step in the ingestion process for RAG?. To convert numerical vector representations back into human-readable tokens for the final output display. To create smaller pieces of text that can be used to add context to the LLM prompts and manage the LLM context window size. To perform a Vector similarity search efficiently by matching the query's vector to the document vectors. To create text embedding for the LLM to process during the final generation step.

Consider this sentence: "You can use multimodal prompts that include both text and images when interacting with an OutSystems AI Agent". Which of the following options is correct?. Provided that underlying AI models configured for the agent specifically support image inputs. All AI models automatically support image inputs regardless of their type. It's not possible, prompts are Strictly limited to text input only. Only if the images have enough resolution and are larger than 20 MB.

Consider a scenario where an Agent receives a user prompt asking from clinics that accept a specific insurance. To process this request, the Agent uses the GetClinics tool and the GetInsurancesByClinic tool once. How many calls are made to the underlying Large Language Model?. 2. 3. 1. 4.

Can single AI Agent be configured to utilize more than one distinct AI Model Simultaneously within its logic?. No, you must create a new Agent for every different AI model you want to introduce into your application. Yes, but only if all the selected AI models originate from the same provider. Yes, an AI Agent can use different AI Models for different purposes within its execution flow. No, an AI Agent is architecturally limited to using exactly one AI model for all its funcionalities.

By default, which component of a conversational Agent actively handles and processes the current conversation's history for contextual responses?. In the StoreMemory Action logic. In the System logs. In the GetGroundingData Action logic. In the System Prompt.

In the Human-in-the-Loop pattern, which specific condition should cause the AI Agent to intentionally interrupt its automated workflow and request input from a human user?. The workflow utilizes the Agent Memory System to retrieve past interactions and relevant domain Knowledge for context. The AI System determines that processing costs for the current multi-step task have exceeded a predefined budgetary constraint. The initial user prompt requires the AI to utilize a tool that is not explicity present in the agent's pre-configured tools list. The AI agent analyzes the current status and reaches a critical juncture or a low-confidence threshold during process.

What is the effect of setting the Max Tokens limit for an AI Agent's Action Calling response?. It sets the maximum number of simultaneous external API calls the agent can make during one turn. It is a security feature that encrypts tokens to ensure truncating the output without generating na error, which helps control costs and verbosity. It limits the final length of the response, potentially truncating the output without generating an error which helps control costs and verbosity. It triggers an immediate error if the agent's internal reasoning exceeds the token count, preventing excessive processing time.

When an AI model uses a Structured Output Definition, what step must be taken if the Agent's call action parameter for the Response Output is not updated to match the structured format?. Explicitly serialize the Structured output into a JSON-formatted string for the Response parameter. Increase the agent's creativity (temperature) setting to force a free-text summary. Completely disable the Structured Output Definition, as it cannot be used if the Response's output parameter is the default string. Set a 5 second mandatory delay to ensure the model has time to process the structure.

Providing the LLM with a small number of examples (typically 2 to 5) to follow a desired output structure or pattern is known as what?. Few-Shot Prompting. Contextual Prompting. Role Prompting. Step-back Prompting.

When comparing Fine-Tuning and RAG, for which scenario is RAG best suited?. Applying implicit domain-specific knowledge. Improving the speed of response for high-traffic applications. Achieving specialized performance, adapted style, and tone. Keeping up with frequently updated knowledge bases or datasets.

Consider a scenario where we need to design an agent that needs to support both function calling (or tool use) and a guaranteed Structured Output definition for its final response. What is the correct approach to achieve this?. Perform two agent calls: the first call executes the necessary tool, and the second call formats that tool’s output into your required structure. Override the system constraint by combining the Action’s instruction and the Structured Output Definition schema within the agent’s initial prompt. Perform just one agent call that executes the necessary tool and also formats the output into your required structure. Use the Action Calling feature and then manually parse the resulting free-text response into the required structure.

In LLM prompting, what defines the AI’s foundational job description, operational mode, or overall behavior throughout an entire conversation or session?. Context Window. Temperature Setting. User Prompt. System Prompt.

Which AI architecture pattern is generally recommended for highly complex, unpredictable tasks, where the sequence of necessary actions is not immediately obvious (e.g., solving an open-ended problem)?. Multi-Agent Workflow, due to its structured and reliable nature. Loop Agent Workflow Pattern, to enforce consistency through repetitive evaluation and refinement cycles. Single Monolithic Agent, to allow the agent to dynamically plan steps and explore paths. Parallel Agent Workflow Pattern, to accelerate execution through simultaneous, but fixed, task handling.

What are the basic units of texts (words, sub-words, or characters) in the context of LLMs?. Weights. Parameters. Tokens. Embeddings.

What is the recommended approach to integrate a proprietary AI Search Service (e.g., a specific vector database) into the ODC ecosystem for use in a RAG-based Agentic Application?. Implement the integration using a regular app, and expose it using a REST API. Then, use that endpoint to set up a Custom AI integration in the ODC Portal. Implement the integration using a Library, and expose it using Server Actions. Then, consume the Server Actions directly in your Agentic app. By configuring the connection of the AI Search Service directly within the Agentic App settings. Implement the integration using a regular app, and expose it via Service Actions. Then, consume the Service Actions directly in your Agentic app.

What is the main purpose of the GetGroundingData Action available in the AgentFlow in an ODC Agentic App?. Retrieve relevant facts from a domain knowledge base to enrich the LLM prompt context. Store conversational history and user preferences in short-term agent memory systems. Define the persona and system rules that guide the LLM’s overall conversational behavior. Invoke external APIs or execute code functions selected autonomously by the AI agent for action.

Consider we have to create an Agent to review the regulations for applying for a loan. These regulations are stored in 100 PDF files, each with 20 pages. What should be done to give the context about these regulations to the Agent?. Configure the Agent’s System Prompt to include a link to the file directory where the 100 PDF files are stored. Store the files in an OutSystems Database and fetch them on the GetGroundingData Action. Store the files in the Azure AI Search Service and use it in the GetGroundingData Action to get context. Store the files in an OutSystems Database and use the Agent AI tools to load the files during the Agent call.

When an AI model uses a Structured Output Definition, what step must be taken in the Agent’s call action parameter for the Response output if it is not updated to match the structured format?. Increase the agent’s creativity (temperature) setting to force a free-text summary. Set a 5-second mandatory delay to ensure the model has time to process the structure. Completely disable the Structured Output Definition, as it cannot be used if the Response’s output parameter is the default string. Explicitly serialize the structured output into a JSON-formatted string for the text Response parameter.

What element is not a key source of contextual information used by an AI model to make decisions about Action Calling or generating a response?. The instructions provided in the System Prompt. The Agent’s configured deployment environment (e.g., Azure or AWS). The text input received from the end-user at runtime. The functional definitions and signatures (metadata) of available Server Actions.

What is the key functional innovation of the Supervisor Agent Pattern that makes it more complex and powerful than a Router Agent or Aggregator Agent Pattern?. It focuses primarily on task concurrency, dividing the workload into independent segments that run simultaneously, followed by combining all results into a single output. It relies on a front-facing agent whose sole purpose is to classify the intent of the user’s input and route that request to a single, pre-determined expert agent for execution. It relies on a centralized agent that actively analyzes the complex request, creates an optimal plan, delegates tasks to various specialized workers, and tracks their real-time progress. It requires all participating agents to continuously refine contributions on a centralized, shared data repository, achieving consensus without explicit message passing.

In the high-level RAG workflow, which operation is performed after the relevant context is retrieved from the Knowledge Base?. The user prompt is converted into vectors. The LLM updates the external Knowledge Base. A hybrid search is performed against the database. The context is added to the prompt and sent to the LLM.

Can a single AI Agent be configured to utilize more than one distinct AI Model simultaneously within its logic?. No, an AI Agent is architecturally limited to using exactly one AI model for all its functionalities. No, you must create a new agent for every different AI model you want to introduce to your application. Yes, but only if all the selected AI models originate from the same provider. Yes, an AI agent can use different AI Models for different purposes within its execution flow.

What does the reason-act loop framework primarily ensure in an AI system?. The agent relies entirely upon a library of preprogrammed, static rules without requiring the consideration of newly acquired environmental data. The agent repeats its last determined action indefinitely until significant state changes happen. The agent communicates its underlying reasoning and decision-making steps to human users before acting. The agent engages in a cyclical process of perception and deliberation to facilitate continuous behavioral refinement and adaptation over time.

When does it make sense to define the Agent’s architecture as Asynchronous?. To ensure strong dependencies are maintained between the Agentic App and the client application. To enable the agent to use external tools like databases or APIs. For long-running model calls, where standard prompts might reach timeouts. To enforce strict real-time user interaction requirements.

Which of the following options is NOT a characteristic of an AI Agent?. The ability to function as an orchestrator by leveraging the LLM’s reasoning to autonomously manage a complex workflow needed to complete a user request. Operating as a proactive problem-solver designed not just to provide an answer, but also to execute actions to real-world tasks. The necessity of agency, which includes breaking down objectives, forming multi-step plans, utilizing external tools, and integrating feedback. The agent’s core distinction is its role as a passive text processor, solely dedicated to optimizing the summarization and translation of documents based on a user’s prompt.

Considering the default service action inputs (UserInput and SessionId) in a typical Agent App, which of the following sentences is true?. The UserInput overrides the System Prompt. The UserInput is used in the System Prompt to define the Agent’s behavior. The SessionId is used to filter the AI model’s responses. The SessionId is used to filter the Memory data.

Which steps are required to make a new AI Model for use within na OutSystems agentic app?. Update the Model Name and Provider parameters directly within th AI Agent's properties in ODC Studio. Install a Forge Connector for the AI Model and add it as public element in your agentic app. Add the new AI Model first within the ODC Portal's managment interface, then add it as a public element in your app. Define the AI Model in a Library, set the model to Public, and add it as a public element in your agentic app.

What is the purpose of the SessionId input variable within the AgentFlow Actions?. To uniquely identifiy and preserve the short-term conversation history across sequential user interactions. To define the specific long-term base partition required for retreving domain facts during RAG. To specify the exact list of tools and external functions that the AI agent is allowed to execute. To provide the Large Language Model with essential persona instructions constraints and operational rules.

In the design of na AI Agent, what type of memory is intended to be a temporary space often implemented as na in-memory object, used specifically to hold information during a single complex task executed through the agent's reasoning loop?. Working Memory (Scratchpad). Long-Term Memory (Persistent). Knowledge Base Memory (RAG). Short-Term Memory (Conversational).

Why is writting clear and detailed Tool descriptions, including input and output parameters a critical step during design of an AI Agent's tool list?. To provide the LLM with sufficent context to accurately select and apply the correct tool in a given situation. To determine the most efficient division of functionality when deciding between single-purpose and multi-purpose tools. To describe the internal implementation details of each tools so the LLM can interpret how the underlying code operates. To guarantee that each tool's inputs and outputs fully comply with the formatting and security standards required by external APIs.

According to ODC reference architecture regarding dependencies what relationship is enforced between na agentic app and a web app?. Only strong references are allowed forcing deployment synchronization. References are completely forbiden to maintain container separation. Only weak references are allowed between web apps and agentic apps. The web app can strongly reference the agentic app, but not the other way around.