Python Institute PCEI: AI Projects and Communication

Topic snapshot

Sample questions

These questions are original IT Mastery practice items aligned to this topic area. They are designed for self-assessment and are not official exam questions.

Question 1

Topic: Block 6: AI Projects, Collaboration, and Communication

A small library wants to reduce repeated phone calls about opening hours. A volunteer proposes the following AI workflow:

Problem value: saves about 2 staff hours per month
Data available: 12 common questions and approved answers
Users: fewer than 30 questions per week
Proposal: train a custom deep-learning chatbot, collect new user data,
and run a 6-month model improvement project
Constraint: no dedicated AI budget or data science staff

What does this project note most strongly indicate?

Options:

• A. The workflow is necessary because all chatbots require deep learning.

• B. The workflow is appropriate because custom training improves every AI project.

• C. The workflow is too small because it lacks a cloud deployment plan.

• D. The workflow is likely too large for the problem scale.

Best answer: D

Explanation: A proportionate AI workflow should match the problem’s value, scale, data, and available resources. Here, the library has a small, predictable question set, low usage, limited benefit, and no dedicated AI team or budget. A custom deep-learning chatbot and long improvement project would likely cost more effort than the problem justifies. A simpler solution, such as a clear FAQ page, scripted responses, or a basic rule-based bot, would better fit the scale.

The key takeaway is to avoid adding AI complexity when a lower-cost solution can meet the same need.

• Deep-learning requirement is a misconception because simple chatbot or FAQ use cases do not automatically require trained neural models.
• Cloud deployment focus fails because deployment architecture is not the limiting issue; value and resource fit are.
• Custom training claim is too broad because custom models are not automatically better for small, predictable tasks.

Question 2

Topic: Block 6: AI Projects, Collaboration, and Communication

A beginner AI project team is preparing a small help-desk dataset before summarizing average open time by priority for resolved tickets. Which data-preparation decision is best supported by the exhibit? Select ONE.

Exhibit: Sample records

ticket_id,priority,resolved,minutes_open
101,High,yes,"45"
102,high,yes,"30"
103,Low,no,"N/A"
104,Medium,yes,"90"

Options:

• A. Standardize priority labels and convert minutes_open to numbers.

• B. Average minutes_open as text because all values are quoted.

• C. Drop priority because its capitalization is inconsistent.

• D. Group by ticket_id because it uniquely identifies each record.

Best answer: A

Explanation: Data preparation often means cleaning values and converting data types before summarizing records. In the exhibit, High and high would create separate groups unless the labels are standardized. The minutes_open values are shown as quoted text, and one unresolved row has N/A, so the field should be converted to a numeric type after selecting the relevant resolved records. These steps make the average by priority meaningful and computable.

The key takeaway is to fix the fields needed for the intended summary, not discard useful attributes or summarize identifiers.

• Text averaging fails because quoted values and N/A are not reliable numeric inputs for an average.
• Dropping priority loses the grouping attribute needed for the requested summary.
• Grouping by ticket ID creates one group per record, which does not answer average open time by priority.

Question 3

Topic: Block 6: AI Projects, Collaboration, and Communication

A beginner team is prototyping a small AI helper that labels clinic messages as urgent or routine. The code runs without errors, but the validation notes show one missed case.

Exhibit: Validation notes

Goal: flag messages needing same-day nurse review.
Current keyword rule: fever, bleeding, chest pain

Message                              True label   Prototype label
"Chest pain after exercise"           urgent       urgent
"Need insulin refill; out tonight"    urgent       routine
"Question about my bill"              routine      routine
"Fever since yesterday"               urgent       urgent

Team note: Developers are unsure which medication/refill phrases require same-day review.

Which collaboration step is best supported by the exhibit? Select ONE.

Options:

• A. Pair program to rewrite the keyword-matching loop

• B. Have a teammate proofread the project README

• C. Replace validation with a larger unlabeled message list

• D. Ask a clinic domain expert to review labels and urgent terms

Best answer: D

Explanation: The core issue is not a Python runtime error or unclear documentation; it is missing domain knowledge. The prototype correctly matches obvious keywords but misses an urgent medication-related message. Because the developers are unsure which medication or refill phrases need same-day review, a clinic domain expert can help confirm labels, identify important terms, and prevent unsafe assumptions. Pair programming or code review is helpful when logic, implementation, or reproducibility is the main concern, but the visible facts point to a domain-definition gap. For small AI tasks, collaboration should match the problem: ask the person who understands the real-world meaning of the data when labels or decision rules are uncertain.

• Code rewrite focus is tempting, but the exhibit says the code runs and the uncertainty is about clinical meaning.
• README proofreading may improve communication, but it does not resolve the missed urgent medical phrase.
• Unlabeled data may add examples, but it will not explain which examples should be urgent without expert labels.

Question 4

Topic: Block 6: AI Projects, Collaboration, and Communication

A community library is planning a small AI feature. Review the project note and choose the best interpretation.

Goal: Route incoming patron emails faster
Input data: Past emails with librarian-assigned topics
Output: Suggested topic for each new email
Constraint: A librarian approves the final response
Stakeholders: Patrons, librarians, library manager

Which statement correctly identifies the project elements?

Options:

• A. It replaces librarians by sending automatic final replies to patrons.

• B. It collects new patron emails only after deployment is complete.

• C. It supports faster email routing using labeled emails and human approval.

• D. It predicts library staffing needs from employee schedules.

Best answer: C

Explanation: For a small AI solution, the project note should be read as a practical map of the work: the goal is the problem to solve, inputs are the data used, outputs are what the system produces, constraints are limits or required controls, and stakeholders are the people affected by or involved in the solution. Here, the AI feature is meant to help route emails faster by suggesting a topic from past labeled examples. The librarian approval requirement means the system is an assistant, not an automatic decision-maker. The stakeholders include both users affected by the service and staff responsible for using or managing it.

• Automatic replies fails because the note says the output is a suggested topic and a librarian approves the final response.
• Staffing prediction fails because the inputs and outputs are about patron emails, not employee schedules.
• Post-deployment collection fails because past labeled emails are already listed as input data for the solution.

Question 5

Topic: Block 6: AI Projects, Collaboration, and Communication

A beginner team is gathering data for a simple AI task that will classify help-desk messages as billing or technical. They want a small dataset that can be loaded easily in Python. Select ONE decision best supported by the exhibit.

Exhibit:

message,category
"My invoice is wrong",billing
"The app crashes on login",technical
"How do I update my card?",billing
"Password reset link fails",technical

Options:

• A. Keep only total counts for each category.

• B. Save the rows as a screenshot image.

• C. Load the messages into a set and discard categories.

• D. Use a CSV with message as input and category as label.

Best answer: D

Explanation: For a simple supervised AI classification task, the dataset should keep each example’s input and its known label in a Python-friendly format. The exhibit already has one text field, message, and one target field, category, arranged like a CSV file. Python can load this into a list of dictionaries, a table, or another simple structure for later cleaning and feature extraction.

Keeping the labels matters because the model needs examples of which messages belong to each category. The key takeaway is to preserve row-level examples in a structured file, not just summaries or visual copies.

• Counts only fail because totals do not preserve individual message examples for training or testing.
• Discarded labels fail because the category is needed as the supervised learning target.
• Screenshot storage fails because an image is harder to load and process than structured text data.

Question 6

Topic: Block 6: AI Projects, Collaboration, and Communication

A beginner AI team tested a simple spam-message classifier for an internal demo. The project report needs a short technical summary for reviewers. It must state the method, data, main result, and conclusion, and it must avoid claiming the model is ready for production.

Project notes:

• Method: logistic regression using word-count features
• Data: 1,000 labeled messages, 70% training and 30% test
• Test accuracy: 88%
• Limitation: only one small internal dataset was tested

Which summary is the best choice?

Options:

• A. A logistic regression model using word-count features was trained on 700 of 1,000 labeled messages and tested on 300 messages, achieving 88% accuracy; the result suggests promise but needs more validation before production use.

• B. The model proves that word-count features cause accurate spam detection across all message sources.

• C. The team used machine learning to classify spam messages, and the model performed well during testing.

• D. The AI spam detector is production-ready because it reached 88% accuracy on the test data and should now replace manual review.

Best answer: A

Explanation: A strong technical summary for an AI project should connect the method, data, measured result, and conclusion in one cautious statement. Here, the method is logistic regression with word-count features, the data is 1,000 labeled messages split into training and test sets, and the measured result is 88% test accuracy. Because the model was tested only on one small internal dataset, the conclusion should not claim production readiness, general performance across all sources, or causal proof. The safest wording is that the result is promising and requires more validation.

• Production-ready claim ignores the stated limitation that only one small internal dataset was tested.
• Vague performance summary misses key technical details such as the method, data split, and exact result.
• Causal overclaim goes beyond the evidence because accuracy on one dataset does not prove broad causation or generalization.

Question 7

Topic: Block 6: AI Projects, Collaboration, and Communication

A team is choosing a ticket-routing approach for a small help desk. Which conclusion is best supported by the project note? Select ONE.

Goal: route each ticket to Billing, Password, or Shipping
Quality target: at least 90% correct on a recent sample
Hosted LLM prompt: 94% correct, estimated 380 USD/month
Offline keyword rules: 92% correct, near-zero extra cost
Output needed: category only, no generated reply

Options:

• A. Train a new deep learning model from scratch.

• B. Use the offline rule-based router first.

• C. Use the hosted LLM because it has the highest score.

• D. Add a generative reply feature to justify the LLM.

Best answer: B

Explanation: Cost-efficient AI project planning starts with the simplest approach that meets the goal. In this note, the task is narrow: assign one of three fixed categories. The offline keyword rules score 92%, which is above the 90% target, and they avoid the recurring hosted LLM cost. The LLM is slightly more accurate, but the extra accuracy is not required by the stated quality target, and the project does not need generated text. A larger or more complex model would add cost and complexity without a clear benefit. The key takeaway is to match solution complexity to the actual requirement.

• Highest score trap fails because 94% is not necessary when 92% already meets the target.
• New model trap fails because training from scratch adds effort for a simple fixed-category task.
• Scope expansion trap fails because generated replies are not part of the stated output need.

Question 8

Topic: Block 6: AI Projects, Collaboration, and Communication

A small team is building a Python-supported AI tool that flags customer support tickets as “urgent” or “normal.” The data analyst reports 90% precision, but the support manager thinks this means the tool finds 90% of all urgent tickets. A pilot decision is due tomorrow, there is no time to retrain the model, and the support manager needs a clear non-technical summary. What is the best collaboration action?

Options:

• A. Proceed with the pilot because 90% precision is strong

• B. Hold a brief alignment meeting with examples and update the project note

• C. Send the Python notebook so the manager can inspect the code

• D. Replace precision with accuracy in the report

Best answer: B

Explanation: The best collaboration action is to translate the technical metric into the stakeholder’s decision context and document the shared understanding. Precision describes how many tickets flagged as urgent are truly urgent; it does not show how many real urgent tickets the tool catches. Because the pilot decision is due tomorrow and retraining is not possible, the team should focus on clear communication, examples, and documented limitations rather than changing the model or hiding the uncertainty. A short meeting plus an updated project note helps both technical and non-technical members use the same definitions and make an informed decision.

• Code handoff fails because reading a notebook does not address the manager’s misunderstanding or need for a non-technical summary.
• Proceeding immediately ignores that the stakeholder may make the pilot decision using the wrong interpretation of precision.
• Changing the metric is misleading because accuracy answers a different question and does not resolve the original misunderstanding.

Question 9

Topic: Block 6: AI Projects, Collaboration, and Communication

A support team is preparing a short technical summary for an internal AI project report. Which summary is best supported by the exhibit? Select ONE.

Exhibit: Evaluation notes

Method: supervised text classifier in Python
Data: 500 labeled support tickets
Features: subject keywords and product category
Evaluation: 100-ticket holdout test set
Results: accuracy 82%; urgent-ticket recall 70%
Error note: 9 of 30 urgent tickets were missed
Intended use: assist triage staff, not replace review

Options:

• A. The model proves that urgent tickets can now be routed automatically because it correctly handled 82% of all cases.

• B. A supervised Python text classifier was tested on held-out ticket data, reached 82% accuracy and 70% urgent recall, and may assist triage staff but still needs human review.

• C. An unsupervised clustering system found hidden ticket groups and should replace manual ticket review.

• D. The project used 500 support tickets, but the results are inconclusive because no evaluation data was provided.

Best answer: B

Explanation: A good technical summary should connect four parts: method, data, results, and conclusion. The exhibit states that the team used a supervised text classifier in Python, trained or developed it with 500 labeled support tickets, and evaluated it on a 100-ticket holdout set. The reported results are 82% accuracy and 70% recall for urgent tickets, with 9 urgent tickets missed. Because missed urgent tickets matter in triage, the conclusion should be cautious: the system may assist staff, but it should not replace human review yet. The best summary reports the visible evidence without claiming full automation or unsupported success.

• Automation claim overstates the result because 70% urgent recall means some urgent tickets were missed.
• Wrong method fails because the exhibit describes supervised classification, not unsupervised clustering.
• Missing evaluation fails because the exhibit explicitly includes a 100-ticket holdout test set and metrics.

Question 10

Topic: Block 6: AI Projects, Collaboration, and Communication

A team is preparing a short update for support managers about an AI model that flags urgent tickets. The managers need to know whether the model is safe to use without extra review.

Exhibit: Evaluation results

Which summary is best supported by the exhibit? Select ONE.

Options:

• A. The model is ready because most tickets are classified correctly.

• B. The model misses many urgent tickets, so extra review is needed.

• C. The data quality problem is duplicate normal tickets.

• D. The model mainly creates too many urgent false alarms.

Best answer: B

Explanation: The key message for support managers should connect the model behavior to the operational risk. In the exhibit, actual urgent tickets are the most important case. The model correctly flags 36 urgent tickets but misses 24 urgent tickets by predicting them as normal. That means 24 of 60 urgent tickets are not caught, or 40% of urgent tickets. Even if many total predictions are correct, the missed urgent tickets are the clearest concern for using the model without human review. A communication summary should emphasize the behavior that affects the audience’s decision, not just a broad accuracy impression.

• Overall correctness is tempting, but it hides the serious missed-urgent-ticket issue.
• False alarms exist, but 10 normal tickets flagged urgent is not the main risk shown.
• Duplicate records are not shown in the exhibit, so that data quality claim is unsupported.

Continue with full practice

Use the Python Institute PCEI Practice Test page for the full IT Mastery practice bank, mixed-topic practice, timed mock exams, explanations, and web/mobile app access.

Try Python Institute PCEI on Web View Python Institute PCEI Practice Test

Related focused pages

• Free Python Institute PCEI Full-Length Practice Exam
• AI Fundamentals
• ML Fundamentals
• Data Handling and Visualization
• Neural Networks and Generative AI
• Responsible AI and Ethics

Free review resource

Read the Python Institute PCEI Cheat Sheet for compact concept review before returning to timed practice.