810-110 Cisco AI Technical Practitioner Practice Questions
Prepare for 810-110 with more than an answer.
- Exam fee
- $150 USD
- Time limit
- 60 minutes
- Questions on the exam
- Not officially published (reports ~40-60)
- Level
- Entry/Associate (Cisco Certified AI Technical Practitioner)
- Valid for
- 3 years
Domains covered on the exam 6
- Generative AI Models20%
- Prompt Engineering15%
- Ethics and Security15%
- Data Research and Analysis10%
- Development and Workflow Automation20%
- Agentic AI20%
- 1
A developer needs an LLM to extract BGP neighbor states from raw CLI output and format them strictly as a specific JSON array. When using a basic instruction prompt, the model occasionally adds conversational text like "Here is the JSON you requested:" which breaks the parsing script.
Which prompting technique is most effective at forcing the model to adhere strictly to the desired JSON output format without conversational filler?
Show answer details
Correct answer: B
Few-shot prompting involves providing the model with a few examples of the exact input and desired output within the prompt. By providing examples that show raw CLI input followed immediately by raw JSON output (with no conversational text), the model recognizes the pattern and accurately mimics the strict formatting constraint.
- 2
A security analyst is using an LLM to investigate a complex network breach. Instead of asking one massive question, the analyst first prompts the model to extract all IP addresses from a log file. In the next prompt, they pass only the extracted IPs and ask the model to identify which belong to known botnets. Finally, they prompt the model to generate a firewall block-list rule for those specific IPs.
What prompting technique is the analyst demonstrating?
Show answer details
Correct answer: A
Chained (or sequential) prompting breaks a complex task into a series of smaller, manageable prompts where the output of one prompt becomes the input for the next. This reduces model confusion and improves accuracy for multi-step reasoning tasks.
sequenceDiagram participant User participant AI as Generative AI User->>AI: Prompt 1: Extract IPs from Logs AI-->>User: Output: List of IPs User->>AI: Prompt 2: Check IPs against Botnet DB AI-->>User: Output: Malicious IPs User->>AI: Prompt 3: Generate Firewall Rule AI-->>User: Output: ACL Config - 3
Which TWO of the following scenarios describe an INDIRECT prompt injection attack? (Select TWO)
Show answer details
Correct answer: A, C
Indirect prompt injection occurs when malicious instructions are placed in third-party data or external sources (like a website or a downloaded document) that the AI model later retrieves and processes. The user interacting with the AI is a victim, not the attacker. Direct injection, conversely, is when the user types the attack directly into the prompt box.
Indirect prompt injection occurs when malicious instructions are placed in third-party data or external sources (like a website or a downloaded document) that the AI model later retrieves and processes. The user interacting with the AI is a victim, not the attacker. Direct injection, conversely, is when the user types the attack directly into the prompt box.
- 4
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?
Show answer details
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.
- 5
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?
Show answer details
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.
- 6
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?
Show answer details
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]
