Hi

Distinguish artificial intelligence, machine learning, deep learning, and generative AI in product discussions.

Explain how probabilistic AI behavior differs from deterministic software behavior.

Classify a product problem as best addressed by rules, predictive machine learning, generative AI, or a non-AI solution.

Identify product failure modes that arise from probabilistic AI outputs.

Map the AI product lifecycle from opportunity discovery through production monitoring.

Frame a user or business problem without prematurely prescribing an AI solution.

Define a measurable desired outcome and a credible current-state baseline for an AI opportunity.

Write an AI use-case hypothesis linking a user problem, an AI capability, a behavior change, and business value.

Distinguish structured, unstructured, and multimodal data in product use cases.

Map the sources, owners, and flow of data required by an AI feature.

Interpret counts, rates, means, medians, percentiles, and distributions in a product data summary.

Distinguish correlation from causation when interpreting product evidence.

Assess a dataset for completeness, accuracy, consistency, timeliness, and representativeness.

Identify features, labels, and prediction targets in a supervised-learning dataset.

Explain the roles of training, validation, test, and live inference data.

Identify data leakage and sampling bias in example AI datasets.

Distinguish supervised learning, unsupervised learning, and reinforcement learning by the feedback each uses.

Match product problems to classification, regression, forecasting, clustering, recommendation, or anomaly-detection tasks.

Explain what happens during model training and model inference.

Define a simple baseline against which an ML model should be compared.

Explain generalization and diagnose underfitting or overfitting from simplified training and validation evidence.

Interpret a model probability and choose a decision threshold based on the costs of false positives and false negatives.

Distinguish global feature importance from a local prediction explanation and state the limitations of each.

Explain how data drift and concept drift can degrade a deployed model.

Distinguish language models, multimodal models, and smaller task-specific models by capability.

Distinguish pretraining, instruction tuning, and inference in the lifecycle of a foundation model.

Explain how tokens and context windows constrain a large language model interaction.

Explain how embeddings represent semantic similarity for search and retrieval.

Write a structured prompt containing a goal, context, examples, constraints, and an output format.

Explain how model settings and prompt wording influence output variability.

Evaluate whether an answer is grounded in supplied evidence and flag unsupported claims.

Assess whether a generative-AI use case has sufficient value, verifiability, and error tolerance.

Decompose a user workflow into tasks, decisions, handoffs, and sources of friction.

Distinguish automation, augmentation, and decision support for a proposed AI intervention.

Identify workflow steps where AI offers an advantage over conventional software.

Formulate user-research questions that test demand for AI assistance and tolerance for errors.

Estimate the time, cost, or loss created by an existing workflow problem.

Score an AI opportunity for user value, data readiness, technical feasibility, and risk.

Prioritize a portfolio of AI opportunities using value, feasibility, and risk evidence.

Write an AI opportunity brief with the problem, users, evidence, assumptions, and expected value.

Distinguish an AI model from the complete product system surrounding it.

Trace an AI request through the interface, orchestration layer, model, data sources, and response handling.

Explain authentication, request and response payloads, rate limits, and errors for an AI inference API.

Choose among synchronous, asynchronous, and batch inference for a product workflow.

Explain how structured outputs and function calling connect a model to software capabilities.

Distinguish a fixed software workflow, a model-directed workflow, and an autonomous agent.

Draw a high-level architecture diagram for an AI-enabled product feature.

Specify AI-system requirements for latency, reliability, scalability, and observability.

Write an AI product requirement that defines the outcome, inputs, expected behavior, and boundaries.

Define the appropriate level of AI autonomy and the points requiring human approval.

Write acceptance criteria for probabilistic behavior using measurable quality and risk thresholds.

Specify fallback and escalation paths for low-confidence or failed AI outputs.

Design an interface that communicates uncertainty without overwhelming the user.

Design a feedback mechanism that captures useful signals for product and model improvement.

Identify edge cases and misuse cases for an AI-enabled workflow.

Create a no-code prototype test plan for an AI interaction.

Distinguish offline model evaluation from online product evaluation.

Interpret a confusion matrix, precision, recall, and F1 score, then explain how classification thresholds change the error tradeoff.

Interpret mean absolute error and root mean squared error for a regression or forecasting use case.

Create a representative evaluation set with important user, data, and risk slices.

Build a rubric for judging the accuracy, relevance, completeness, and safety of generated responses.

Plan a human evaluation process for outputs that require subjective judgment.

Select model, user, business, operational, and safety metrics for an AI feature.

Design an online experiment with success criteria and guardrail metrics for an AI feature.

Explain how retrieval-augmented generation grounds model responses in external knowledge.

Map the ingestion, chunking, embedding, indexing, retrieval, and generation stages of a RAG system.

Select chunking and metadata choices for a document-retrieval use case.

Distinguish retrieval failures from generation failures in a grounded assistant.

Choose among prompt engineering, RAG, and fine-tuning for a product requirement.

Explain how tool calling allows a model to take actions in external systems.

Define the tools, state, stopping conditions, and human checkpoints for an agentic workflow.

Assess whether an agentic approach is justified by task complexity, value, and risk.

Categorize potential AI harms affecting users, organizations, and society.

Create an AI risk register that scores likelihood, severity, detectability, and mitigation.

Assess model performance across relevant user or operational slices for potential unfairness.

Apply data minimization, consent, retention, and access-control principles to an AI feature.

Identify prompt injection, data exfiltration, insecure tool use, and sensitive-output risks in an LLM product.

Define preventive guardrails and adversarial tests for a high-risk AI workflow.

Specify human oversight, contestability, and appeal mechanisms for consequential AI decisions.

Create a governance checklist covering ownership, documentation, policy or regulatory review, monitoring, and incident response.

Compare build, buy, and partner options for an AI product capability.

Create a vendor evaluation scorecard covering quality, data handling, integration, cost, and lock-in.

Scope an AI minimum viable product around the riskiest assumptions.

Estimate per-task AI costs and test whether expected value supports the unit economics.

Define responsibilities across product, design, data, engineering, domain, security, legal, and operations roles.

Choose a shadow, canary, phased, or full rollout strategy for an AI feature.

Design a production dashboard for data quality, model quality, user outcomes, reliability, safety, and cost.

Use production errors and drift signals to prioritize rollback, retraining, prompt changes, or workflow changes.

Frame a late-shipment risk prediction feature for a logistics platform, including users, data, baseline, and success metric.

Define evaluation and intervention requirements for a demand-forecasting use case.

Design a human-in-the-loop document extraction workflow for invoices or shipping documents.

Design a grounded operations copilot that answers questions from logistics policies and shipment records.

Define an anomaly-detection product for operational or transaction monitoring.

Specify the interaction design, fallback logic, and evaluation criteria for a generative customer-support copilot.

Specify safety and approval controls for an agent that resolves a multi-step operational exception.

Compare the studio use cases and produce an evidence-based sequence for testing the strongest opportunities.