FreeLevel 3 · Professional

Generative AI Engineering

This is not a course on using chatbots. It teaches you to build AI features that a business can trust. You'll design the support assistant for Paystream, a Nigerian mobile wallet, using its help centre, 900 labelled support tickets and graded answers from two assistant versions. You'll learn how models work and why they make confident mistakes, call them safely from Python with keys kept out of code, write prompts as versioned specifications, and validate every output against a schema. You'll compare a small model, a large model and a classic text classifier on the same tickets, build retrieval and measure its hit rate, show how grounding with citations and refusals changes answer quality, and check an LLM judge against human grades before trusting it. Finally, you'll redact personal data, defend against prompt injection, cost the options, and write a launch recommendation the evidence supports. Every number in the lessons comes from running the code; live model calls are optional and shown with the Anthropic SDK.

About 7 hours
10 modules
1 badge
Final project
Optional verified certificate

Curriculum

01How Large Language Models WorkWhat a large language model actually does (predict the next token), what tokens and context windows are, why models sound confident when they're wrong, and what that means for anyone building with them.15 min
02Calling a Model from CodeCall an LLM from Python through an API, keep your key safe, control the output with system prompts, temperature and token limits, and estimate what a feature will cost before you build it.15 min
03Prompts as EngineeringWrite prompts the way engineers write specifications (role, task, context, rules, examples and output format), keep untrusted input clearly separated, and version and test prompts like code.25 min
04Structured Outputs and ValidationGet model outputs your code can use, by asking for JSON, validating every response against a schema, and handling the ones that fail instead of trusting them blindly.25 min
05Evaluating an LLM ClassifierMeasure an LLM's classifications against human labels, compare a small model, a large model and a classic machine learning baseline, and find where each one fails.25 min
06Embeddings and SearchTurn text into vectors, find the most relevant documents for a question, measure retrieval with hit rate, and set a threshold for questions the documents can't answer.15 min
07Retrieval-Augmented GenerationBuild the prompt that turns retrieved articles into a grounded answer with citations and honest refusals, and measure how much retrieval improves answers over a model answering from memory.20 min
08Evaluating Generated AnswersBuild an evaluation for free-text answers from human grades, automatic checks and an LLM judge, and find out how far the automated judge can be trusted before relying on it.25 min
09Safety, Privacy and CostRemove personal data before text reaches a model, defend against prompt injection, keep people in charge of high-stakes cases, and choose a model by weighing accuracy against cost.25 min
10Final ProjectPlan your final project, a support assistant prototype with ticket triage and grounded answers, evaluated honestly and made safe, and set up the evaluation harness first.20 min
Final assessmentPass it (60% or more) to earn the Generative AI Engineering course completion badge.

Skills you'll learn

How LLMs work: tokens, context windows and hallucination
Calling models from Python with keys kept secret
Prompts as versioned specifications
Structured outputs and schema validation
Evaluating LLM classifiers against labels and baselines
Retrieval, hit rate and refusal thresholds
Retrieval-augmented generation with citations
LLM-as-judge, checked against human grades
Privacy, prompt injection, cost and launch decisions

Before you start

Python for Data Analytics, or comfort with pandas
Machine Learning Fundamentals is helpful, for train and test splits and recall

Final project

Paystream support assistant: prototype and evaluation

Badges and certificate

Earn a free course completion badge with its own credential ID and public page you can share. Learning and badges are always free.

After you finish, you can choose to get an official verified PDF certificate. It's optional. How it works

Loading…

Generative AI Engineering

About 7 hours

10 modules

1 badge

Final project

Optional verified certificate

Curriculum

01How Large Language Models WorkWhat a large language model actually does (predict the next token), what tokens and context windows are, why models sound confident when they're wrong, and what that means for anyone building with them.15 min

02Calling a Model from CodeCall an LLM from Python through an API, keep your key safe, control the output with system prompts, temperature and token limits, and estimate what a feature will cost before you build it.15 min

03Prompts as EngineeringWrite prompts the way engineers write specifications (role, task, context, rules, examples and output format), keep untrusted input clearly separated, and version and test prompts like code.25 min

04Structured Outputs and ValidationGet model outputs your code can use, by asking for JSON, validating every response against a schema, and handling the ones that fail instead of trusting them blindly.25 min

05Evaluating an LLM ClassifierMeasure an LLM's classifications against human labels, compare a small model, a large model and a classic machine learning baseline, and find where each one fails.25 min

06Embeddings and SearchTurn text into vectors, find the most relevant documents for a question, measure retrieval with hit rate, and set a threshold for questions the documents can't answer.15 min

07Retrieval-Augmented GenerationBuild the prompt that turns retrieved articles into a grounded answer with citations and honest refusals, and measure how much retrieval improves answers over a model answering from memory.20 min

08Evaluating Generated AnswersBuild an evaluation for free-text answers from human grades, automatic checks and an LLM judge, and find out how far the automated judge can be trusted before relying on it.25 min

09Safety, Privacy and CostRemove personal data before text reaches a model, defend against prompt injection, keep people in charge of high-stakes cases, and choose a model by weighing accuracy against cost.25 min

10Final ProjectPlan your final project, a support assistant prototype with ticket triage and grounded answers, evaluated honestly and made safe, and set up the evaluation harness first.20 min

Final assessmentPass it (60% or more) to earn the Generative AI Engineering course completion badge.

Skills you'll learn

How LLMs work: tokens, context windows and hallucination

Calling models from Python with keys kept secret

Prompts as versioned specifications

Structured outputs and schema validation

Evaluating LLM classifiers against labels and baselines

Retrieval, hit rate and refusal thresholds

Retrieval-augmented generation with citations

LLM-as-judge, checked against human grades

Privacy, prompt injection, cost and launch decisions