The Library

Turns a working LLM call into a service that survives real traffic: retries, queues, cost and latency control, fallback to a second provider, caching and observability. Both exercises call the real OpenAI API.

For Python developers with one LLM API call already behind them (course 1 of the From 0 to AI Engineer series) · 9 chapters · 21 min · 2 exercises

Builds a retrieval pipeline piece by piece: chunking, embeddings, cosine similarity and grounded answering, testing each part before trusting it. Deliberately no vector database, and chapter 5 explains what one solves and when you actually need it.

For anyone who finished course 1, or knows the equivalent (course 2 of the From 0 to AI Engineer series) · 9 chapters · 34 min · 2 exercises

Goes from judging one answer to evaluating a whole system: golden datasets, aggregate metrics that catch regressions between prompt or model versions, structured logging, p50/p95/p99, drift detection, and alerts that do not cause fatigue. Almost all pure logic, so both exercises run without an API key.

For anyone who finished course 1 and has tried LLM-as-a-judge (course 3 of the From 0 to AI Engineer series) · 9 chapters · 34 min · 2 exercises

Covers the think/act/observe loop and function calling with real tools, when a deterministic workflow is enough instead of an agent, and the step limits, action validation, memory and cost controls that keep a multi-step agent from running away. Both exercises run without an API key.

For anyone who finished courses 1 to 3 (course 4, the last of the From 0 to AI Engineer series) · 9 chapters · 33 min · 2 exercises

Starts from zero: variables, conditionals, loops and functions, test types and levels, the defect lifecycle, comparing expected against actual data, basic Git, and writing a bug report someone else can follow. Includes a quiz, a code challenge graded in your browser, and a certificate.

For people who have never written code and want to get in through testing (course 1 of the Python for QA series) · 9 chapters · 26 min · 5 exercises

Moves to data that lives outside your own machine: the requests library and JSON, HTTP status codes and assertions, basic SQL (SELECT, WHERE, JOIN), connecting Python to a database, checking the database against the application, and organising it all with pytest, reports and logging. Includes a quiz, a browser-graded code challenge, and a certificate.

For anyone who finished course 1, or knows the equivalent (course 2 of the Python for QA series) · 10 chapters · 24 min · 6 exercises

Takes automation up to what users actually see: UI testing with Playwright and the Page Object Model, pytest fixtures, parametrisation and markers, Docker and a real CI/CD pipeline on GitHub Actions, plus load testing with Locust, mocking, and fake data with Faker. Includes a quiz, a browser-graded code challenge, and a certificate.

For anyone who finished courses 1 and 2 (course 3 of the Python for QA series) · 12 chapters · 25 min · 5 exercises

Brings LLMs into the test suite: generating test cases and summarising failures with the OpenAI API, self-healing UI tests, LLM-as-a-judge, catching hallucinations, and the LangChain library. Includes a quiz, a browser-graded code challenge and a certificate. The exercises call the real OpenAI API, so you need your own key and a few cents of usage.

For anyone who finished course 2 and will have to test an AI product (course 4, the last of the Python for QA series) · 9 chapters · 28 min · 3 exercises

Walks through job interviews in any field: what they ask and why, how to answer with the STAR method, what an employer is not allowed to ask, and a log for every interview you sit. Ends with example answers for customer service, sales, admin, logistics, hospitality, health, construction, transport, education and tech, plus first job and career change. 25 chapters.

For anyone with an interview coming up, whatever the job · Interview Log

The legal chapter is Portuguese labour law (Código do Trabalho, articles 16 and 17), and the pay and notice-period sections follow Portuguese practice, including the fourteen-month salary year.

Covers how to tell a well-told answer from a proven one, how to build a scoring grid before the conversation, why one interviewer alone is not enough, and what you are not allowed to ask. 9 chapters, about 13 minutes.

For people who ended up interviewing candidates without ever being trained to · Interview Scorecard

Chapter 8, on the questions you cannot put to a candidate, is Portuguese labour law.

Sorts facts from opinions, then shows how to prepare your own side of the review, how to keep evidence through the year without sounding vain, and what to say when you disagree with the result. 8 chapters, about 10 minutes.

For the person being reviewed, not the person doing the reviewing · Evidence Log

Lays out what a manager actually does once the promotion arrives without a manual: coaching without the jargon (GROW), goals that do not get forgotten (SMART), feedback people take seriously, and what to do when someone stops improving. 10 chapters, about 16 minutes.

For first-time managers, promoted with no warning · 1:1 Log

Deals with friction between teams rather than inside one: the five ways people react to a disagreement, why the problem is rarely the person and almost always a different objective, how to disagree without escalating, and when escalating is the right call. 8 chapters, about 9 minutes.

For anyone who depends on people who do not report to them · Alignment Log

Looks at what happens after the room empties: why the end-of-session survey measures the wrong thing, why the field's most quoted statistic was never actually established, what to measure instead, and what to do with a difficult group. 10 chapters, about 15 minutes.

For trainers, and for managers who have to onboard new people · Training Transfer Plan

Explains why technical skill is no substitute for adoption: Sean Ellis's 40% test, the B=MAP model of friction, and real cases from Google Wave to Dropbox. Each of the 12 chapters closes with a question to answer, one action for the week, and a way to measure whether anything moved.

For people building a product · Idea Validator

Introduces AI for day-to-day work without the hype: the vocabulary with no jargon, the shift in mindset that changes the answers you get, the three mistakes beginners make, and a place to keep the prompts that work. 10 chapters, based on three years of daily use rather than research, with sources cited where research confirms or contradicts it.

For anyone who never really used AI, or tried once and gave up · Prompt Library

Shows why an AI's confidence is not evidence, why one reviewer is never enough even when it is an agent, and why a test proves what an explanation only makes convincing. 8 chapters, two of them real cases where the check caught something about to ship.

For people building with AI who need to trust the result · Verification Log

The full guide is published in English, not just this description.

Digs past the first plausible explanation: the five whys, the Ishikawa diagram, how to tell that you have actually reached the cause, and what to do when the cause turns out to be you or your own process. 8 chapters.

For anyone with a problem that keeps coming back · Root Cause Log

A first-person account: why a diploma left untouched for fifteen years does not count as a starting point, why starting from what hurts beats starting from theory, and the difference between a script and a system. 8 chapters, one person's route rather than a method tested on anyone else.

For people who did not come through a straight technical path · Progress Log

Separates creating value from capturing it: why the market buys results rather than capability, a job seen as the same sale made over and over, the blocks that are not technical, and then positioning, offer and price. 9 chapters, written from inside the problem rather than from the far side of a result.

For people who are good at the work and still do not invoice · Offer Log

An essay rather than a guide, with no method and no sources: where you actually stand against what you said you wanted a year ago, how much of what you are missing is really missing, and the smallest possible version of the thing you keep putting off. About seven minutes, and it only works if you answer the questions.

For anyone who has been postponing something

A two-column CV template ready to print to PDF, plus a single-column version for the ATS application portals that scramble column layouts. The preview updates as you type, and nothing leaves your browser.

For anyone who needs a CV today

The form and the CV's section headings are in Portuguese (Resumo, Experiência, Formação); the text you type appears exactly as you write it.

Converts a photo of your pen signature into animated SVG, either tracing the ink exactly as it sits on the paper or reconstructing the pen path by skeletonization and redrawing it stroke by stroke. A photo goes in, an SVG comes out, and all of it runs in your browser, so the image never leaves your computer and is never stored.

For anyone who wants their own signature as a file

The closing caution about publishing a signature is written around Portuguese law.