Cisco AI Technical Practitioner Free Sample Questions
Description
Covers generative AI model families and hosting trade-offs, prompt engineering techniques and injection risks, responsible AI and data privacy, AI-assisted data analysis, and agentic AI with MCP.
12 free sample questions150 in the full practice test
Generative AI Models · Describe major generative AI model families and common use cases
A network automation team needs to generate Python scripts to configure OSPF on Cisco Catalyst 9300 switches. They are evaluating different generative AI model families for this task. Which model family is fundamentally designed and most appropriate for generating this configuration code?
Answer and explanation
Correct answer: B
Large Language Models (LLMs) rely on transformer architectures that excel at understanding and generating sequential data, making them ideal for text summarization, content creation, and code generation (like Python scripts for Cisco devices). Diffusion models are primarily used for image/audio generation, and GANs are typically used for synthesizing realistic media rather than logical code structures.
Q2Question 2
Intermediate
Generative AI Models · Compare model hosting options and their trade-offs
A healthcare provider is deploying an AI assistant to analyze patient telemetry data originating from Cisco Meraki IoT sensors. Due to strict HIPAA compliance, patient data cannot leave the corporate network boundary. However, the organization has limited capital budget for GPU hardware. Which hosting strategy best balances these trade-offs?
Answer and explanation
Correct answer: D
The strict constraint that data 'cannot leave the corporate network boundary' mandates a locally hosted (on-premises) model to ensure data privacy and compliance. To address the 'limited capital budget for GPU hardware' constraint, using a smaller, quantized model allows inference to run efficiently on lower-end hardware or even CPUs, successfully balancing the privacy and cost trade-offs.
Q3Question 3
Advanced
Generative AI Models · Explain role of context windows, token limits and response management
Enterprise architecture team at a financial firm is building a system to analyze 500-page regulatory compliance documents. They must decide between using a model with a massive 1-million token context window or implementing a Retrieval-Augmented Generation (RAG) architecture with a smaller context window.
The documents contain highly specific, dense legal clauses. The business requires near real-time answers (low latency) and minimal API invocation costs per query. They also require the ability to trace the exact paragraph source of the AI's answer.
Based on these constraints, which architectural approach is optimal and why?
Answer and explanation
Correct answer: A
RAG is the optimal choice here. Feeding a 500-page document into a massive context window for every query results in extremely high token costs and high latency due to the attention mechanism processing the entire document each time. RAG mitigates this by retrieving only the relevant chunks (lowering cost and latency) and inherently provides traceability by showing exactly which retrieved chunks were used to generate the answer.
quadrantChart
title RAG vs Long Context Trade-offs
x-axis Low Cost --> High Cost
y-axis High Latency --> Low Latency
quadrant-1 Unfavorable
quadrant-2 Ideal
quadrant-3 Suboptimal
quadrant-4 Acceptable
Long Context Window: [0.85, 0.80]
RAG Architecture: [0.20, 0.30]
Q4Question 4
IntermediateMultiple answers
Generative AI Models · Understand model selection in AI model hubs and repositories for appropriate use-cases
When browsing AI model hubs like Hugging Face to select a model for analyzing network topology diagrams (images) and producing a written summary of potential single points of failure, which TWO model characteristics are strictly required? (Select TWO)
Answer and explanation
Correct answers: D, E
To analyze network topology diagrams (images) and output text, the model must be multimodal (specifically, a Vision-Language Model). Furthermore, identifying single points of failure requires logical deduction and architectural understanding, meaning the model must possess strong reasoning capabilities rather than just basic image captioning.
Q5Question 5
Intermediate
Generative AI Models · Describe Retrieval Augmented Generation (RAG) and role of embeddings and vector databases
In a Retrieval-Augmented Generation (RAG) architecture built to query Cisco Nexus configuration guides, what is the primary role of the vector database during the retrieval phase?
Answer and explanation
Correct answer: A
In a RAG system, the vector database stores embeddings (mathematical representations) of document chunks. When a user submits a query, it is also converted into an embedding. The vector database performs semantic search by calculating the mathematical distance (e.g., cosine similarity) between the query vector and the stored vectors to return the most relevant context chunks.
Generative AI Models · Explain role of context windows, token limits and response management
A data engineer is optimizing a RAG pipeline that indexes Cisco design guides. They notice that the LLM frequently hallucinates details when answering complex architectural questions, even though the correct documents are being retrieved. Upon inspection, the chunk size is set to 150 tokens with no overlap.
How does this chunking strategy negatively impact the model's response management?
Answer and explanation
Correct answer: C
Chunking strategy directly impacts what context is injected into the prompt. A very small chunk size (150 tokens) with no overlap often cuts sentences or concepts in half. When the vector DB retrieves these fragmented chunks, the LLM lacks the surrounding context required to formulate a complete, accurate answer, leading to hallucinations as it tries to fill in the missing information.
Register free for 6 more questions
Or unlock all 150 810-110 questions with explanations, timed mode and flashcards.
