Luiz Trajano
Porto, PortugalOpen to data analyst roles and freelance projectsRemote or hybrid · on-site in Northern PortugalSQL · Python · A/B testing · Dashboards
Final project · cohorts & hypothesis testing · 2026
Only 2.01% of 58,139 first-time buyers came back within 180 days. A late first delivery lifts 1–3 ratings from 18% to 76%, and five hypotheses tested with Holm and power.
A late first delivery turns 75.63% of ratings into 1 to 3, against 18.17% on time, and customers who rate 1 come back less: 1.68% against 2.10%.
Machine learning · 2026
A model that finds 83% of the customers about to leave a gym chain, a leakage test on the suspect feature, and five customer groups whose churn runs from 2.2% to 52.6%.
Two customer groups hold 828 of the 1,061 people who left: 52.6% churn in the low-engagement group against 2.2% on annual contracts.
Tableau dashboard · 2026
A Tableau Public dashboard over 12,343 daily records in five countries, with date and country filters driving all four views, and the three caveats that change how it reads.
Japan's share of trending video-days goes from 10.8% to 17.0% once the comparison starts where every country has data.
Hey!
I'm Luiz, a Data Analyst in Porto, Portugal, originally from Caruaru, in the Agreste region of Pernambuco, Brazil. I work with SQL, Python, A/B testing and Tableau, and have ten public projects from the TripleTen Data Analyst program. Before data, I worked logistics and operations at a supply warehouse for a supermarket chain and played basketball at a semi-professional level. Both taught me the instinct I still rely on: read the game, spot the pattern, help the team decide.
That instinct found a technical home early. A systems-development diploma back in Caruaru gave me logic and databases before I knew what to do with them, and four months on a study exchange in Stratford, Ontario turned classroom English into the fluency I use every day. Data was the natural next step.
I've completed the TripleTen Data Analyst program, with ten end-to-end public projects, from notebook to production: SQL, A/B tests, dashboards and a final project on customer repurchase, each one answering a real business question. Python is part of my daily work, and Excel and Power BI are my next step, in self-study. Claude and Claude Code are part of that workflow every day, giving support and speeding up the process, the quickest route I've found from question to shipped analysis. The analyses and the answers always come from my own conclusions.
Escola Técnica Estadual Ministro Fernando Lyra · Caruaru, Brazil
High-school track integrated with a technical diploma (2017–2019, 5,400 hours): programming logic, databases, web, desktop and mobile development — the base I later carried into data.
Issued 6 May 2020 · Nº SISTEC 52674/71988122CM
Study exchange · Stratford, Ontario, Canada
Four months living and studying in Canada, in English full time, Feb–Jun 2019: not classroom English, but everyday conversation, teamwork and professional settings.
Issued 14 Jun 2019
TripleTen · Dec 2025 – Jul 2026
Fourteen modules and 465–470 hours, each project approved by a TripleTen reviewer: SQL with CTEs and window functions, Python for EDA, hypothesis testing (t-tests, Mann-Whitney, chi-square and z-tests with Bonferroni and Holm corrections), cohorts and unit economics, funnels and A/B tests, Tableau dashboards, machine learning with scikit-learn and a final project on customer repurchase. Every project is public on GitHub.
Issued 14 Oct 2026 · Nº 202603DAEN001477
From idea to insight. Clean, organised datasets built for clear presentation and better decisions.
Pandas · Business insight · Report
Tableau · Streamlit · Plotly
SciPy · Non-parametric statistics · Prioritisation
Quick diagnosis · Recommendation · Automation
Data manipulation and analysis
Pandas, NumPy and SQL are my base tools: cleaning, joining and querying data of any size, turning raw information into something ready to analyse.
Cleaning and interactive presentation
I take messy, disorganised data and make it trustworthy, then turn the result into interactive dashboards anyone can explore and understand.
Claude & Copilot, after the work is done
The analysis, the code and the conclusions are mine. Once I finish, I use Claude and Copilot as a second reader: checking numbers, pointing out mistakes and answering syntax questions. When a suggestion holds up, I fix it and explain the change.
Fluency applied at work
I write documentation, present results and talk to clients in English without friction. The fluency I brought back from the exchange in Canada is a daily work tool now, not just a memory from the trip.
seaborn & matplotlib
Boxplots, distributions and comparison charts across sprints 6 to 14: the plots that carried each finding, from platform sales to ride durations under bad weather.
Dashboards on Tableau Public
Sprint 12: a published dashboard with four views, date and country filters applied to every sheet, percent-of-total calculations checked against pandas, and the data caveats written into the dashboard itself. Sprint 14: a repurchase dashboard filtered by month, state, rating and delivery, reading the same numbers as the notebook.
scikit-learn
Sprint 13: logistic regression and a random forest compared against a baseline, a stratified split, precision and recall beyond accuracy, a leakage test that drops the suspect feature, and K-Means with a dendrogram and silhouette to segment 4,000 customers.
SciPy
Welch and Student t-tests, Mann-Whitney on skewed revenue, variance checks before every test, and a hand-written z-test for proportions with a Bonferroni correction across 20 tests. In sprint 14, chi-square before the pairwise z-tests, a Holm correction, 95% intervals and the smallest difference each test can detect at 80% power, so a "no difference" is read alongside its power. Sprints 6, 7, 9, 11 and 14, each with the test choice justified in the notebook.
Decomposition & cohorts
Sprint 14: the business question split into five hypotheses and their KPIs before touching the data, a cohort where every customer had the same 180 days to come back, and the main results re-run with 90- and 365-day windows.
BeautifulSoup & requests
Scraped Chicago weather records from an HTML table and joined them to trip data by timestamp for the Zuber hypothesis test in sprint 7.
Versioned portfolio
Every sprint lives in its own public repository, and this site ships from a Git-connected pipeline: branch, preview, merge, production.
Self-study · next step
Business dashboards and DAX are my next step after TripleTen. This card fills in when the first real report is done.
PostgreSQL
Sprint 14: CTEs that aggregate each table before the join, window functions (RANK and SUM OVER) for rankings and shares, and LEFT JOIN … IS NULL integrity checks across a five-table books database, building on the PostgreSQL work from sprint 7.
Self-study · next step
Pivot tables, lookups and analysis workflows for the teams that live in spreadsheets. Next on my self-study list.
CAR
format of the final presentation
Every project ends in a recommendation someone without a stats background can act on. In sprint 14 the deliverable was a presentation in the CAR format (context, action, result) and a dashboard that reads the same numbers as the notebook.
7
reviewer points, each answered
Each project went through a TripleTen reviewer before approval. In sprint 14 the reviewer left seven points on the analysis plan, and the notebook records what I did with each one and where.
~19
extra customers: the size of the effect
I report how big an effect is, not just whether it passes a test. In sprint 14 the delivery delay was significant, yet worth about 19 extra returning customers, and the recommendation says so.
14
modules finished
I research a method before I use it and write down why I chose it: test power, the Holm correction, cohort windows. Fourteen modules finished, and every project is public.
Basketball
semi-pro, then warehouse operations
Semi-professional basketball and warehouse operations taught me to read a situation fast, play my role and help the team decide.
3
caveats nobody asked for
I question the data before I trust it. In sprint 12 I found three caveats nobody asked for: the metric counted video-days, not videos; Japan had a shorter collection window; and eight days had no data. In sprint 14 I caught a second A/B test running on the same users.
2
changes for group B
When a question is ambiguous, I pin it down before answering. "Last week" with no reference date became a declared window in sprint 12, and in sprint 14 I wrote down that the brief named two different changes for group B before reading any result.
+1
optional validation notebook
In sprint 12 I added an optional validation notebook that checked every number in the dashboard against pandas, and the reviewer called it the best part of the package. In sprint 14 I ran the robustness checks the reviewer had only suggested.
Contact
Have a project, or need a second pair of eyes on your data? Fill in the form and I will get back to you.
Other regions of Portugal: open to interviews and to discuss.