Generative AI Without Advanced Math

Store a model API key in an environment variable without exposing it in Python source code.

Configure and verify an API usage limit or billing alert that keeps paid experiments within an $8 budget.

Send a single request to a large language model from Python and print its response.

Construct a prompt containing a clear task, relevant context, and explicit constraints.

Test a prompt against a small set of examples and record where its responses fail.

Implement a command-line AI assistant that preserves conversation history.

Distinguish rule-based software, traditional machine learning, and generative AI from their inputs and outputs.

Compute and interpret ratios and percentages from model-evaluation counts using Python.

Calculate absolute differences between predicted and actual values using Python.

Calculate and interpret the arithmetic mean of a small dataset with Python.

Interpret a probability score as a value between 0 and 1 and as a percentage.

Plot points on a 2D chart with Python and interpret coordinates and numerical scales.

Generate a table of outputs from a one-input Python function and describe how its output changes.

Use Python to calculate the distance between two points and use smaller distances to rank proximity.

Formulate one practical classification question and one practical regression question based on the type of output required.

Inspect prepared regression and classification datasets and separate each into input features and a target variable.

Split each dataset into training and test sets and explain the purpose of each set.

Fit a linear regression model with scikit-learn and generate predictions.

Calculate mean absolute error from regression predictions and actual values.

Fit a decision-tree classifier with scikit-learn and generate predicted classes.

Construct and interpret an accuracy score and confusion matrix for a classifier.

Compare training and test performance to identify basic evidence of overfitting.

Explain how token-by-token text generation can produce fluent but incorrect answers.

Measure a prompt's token usage and shorten it to fit within a model's context limit.

Create a reusable prompt template whose values are supplied by Python variables.

Request structured JSON output from a model and validate the returned fields.

Handle a model API timeout or rate-limit error without crashing the application.

Add an instruction that makes the model state when supplied evidence is insufficient.

Build a simple Streamlit interface for a generative-AI text-processing application.

Evaluate the application on a fixed set of cases and revise one recurring failure.

Distinguish information learned during model training from context supplied at request time.

Extract and clean text from a document while preserving source or page metadata.

Split document text into source-labeled chunks suitable for retrieval.

Generate embeddings for short texts and use a similarity score to rank which texts are semantically closest.

Retrieve the most relevant document chunks for a question using vector similarity search.

Construct a prompt that requires an answer and citations from retrieved context.

Make the application decline to answer when retrieved evidence is insufficient.

Evaluate retrieval relevance and answer faithfulness with a small question set.

Distinguish a deterministic workflow, a tool-calling assistant, and an autonomous agent.

Describe a Python function as an AI-callable tool using a structured argument schema.

Make a model select an appropriate tool for a user request.

Validate model-produced tool arguments before executing a Python function.

Execute only an allow-listed tool and return its result to the model.

Maintain explicit state across the steps of a multi-step AI workflow.

Limit tool usage with step limits, timeouts, and confirmation rules.

Test a tool-using workflow against successful, invalid, and unsafe requests.

Package a selected project as a reproducible Python repository with declared dependencies.

Configure deployment secrets without committing credentials to source control.

Deploy the selected project as a small web application on a hosted service.

Log request outcomes, latency, and token usage for a deployed AI application.

Estimate monthly model cost from request volume and average token usage.

Test the deployed application for basic prompt-injection and sensitive-data leakage risks.

Write a concise portfolio case study describing the problem, design choices, evaluation, and limitations.

Compare firsthand project tasks with AI role profiles and select a focused next learning path.