I build the layers between marketing activity and the money it produces: measurement that can be trusted, analysis that survives being questioned, and models that end in a decision someone can act on.
Around five years across paid media, SaaS growth and analytics, mostly in businesses where the marketing number and the finance number had to agree. What I am useful for is the join: connecting campaign behaviour to CAC, LTV, payback and contribution margin, then building the thing that computes it instead of stopping at the recommendation.
Every repository answers one question in a growth system: validate that the signal is real, scale what works, retain the value it creates. A fourth track holds applied research and products.
Every case follows the same arc: Context → Problem → Strategy → Result, followed by its limits and the next move.
| Problem | Result | |
|---|---|---|
| marketing-mix-modeling | How much revenue did each channel cause? On real data, an MMM cannot be scored. | Fitted against a known process: perfect ranking on every channel with signal. A pre-fit signal-to-noise check flags the one it cannot measure (+636% ROI error). |
| ab-testing-toolkit | Can this experiment answer its question, and what do we do with the answer? | Separates no effect from no power. CUPED cuts variance 75%, worth 4x the traffic. Calibrated on 2,000 simulated A/A tests. |
| tracking-attribution-lab | Can this CPA be trusted? | Five named event states and SQL checks that catch loss before reporting. |
| Problem | Result | |
|---|---|---|
| unit-economics-olist | Which acquisition channels pay for themselves in a marketplace? | 97% of customers buy once, so CAC must clear on the first order. Paid channels consume 71% of a R$ 16.13 margin ceiling. |
| lead-scoring-api | Who should a call centre call first, before anyone dials? | The usual 0.954 AUC becomes 0.640 once two leaks are removed. What survives: a 1.54x lift, and 30% of capacity reaching 46% of conversions. FastAPI + Docker. |
| growth-analytics-warehouse | Where does paid media pay back, and can the data be trusted? | End-to-end synthetic platform: snowflake warehouse, 68 quality checks, SQL catalog, model and app. Only the US clears 3x LTV/CAC. |
| paid-media-budget-optimizer | Where does the next unit of budget go? | A transparent score where headroom, not efficiency alone, decides growth. |
| marketing-analytics-portfolio | How does a report become a decision? | One KPI layer and one case format that ends in a decision and a limitation. |
| Problem | Result | |
|---|---|---|
| clv-cohort-prediction | Who deserves retention budget, predicted before the spend? | BG/NBD + Gamma-Gamma beats LightGBM on every metric (MAE £484 vs £645). The top decile captures 51.2% of holdout value. |
| churn-cost-sensitive | Who gets a retention offer that costs money and works only sometimes? | Break-even churn probability ranges from 2.7% to 19.6%. Thresholds from retention economics recover £10,285, 21% of achievable value. |
| Result | |
|---|---|
| carbon | Exon/intron classification in human DNA with a Bi-LSTM, 99.80% test accuracy against three controlled baselines. Paper accepted at an international bioinformatics conference. |
| visppy-cv | Computer vision for physical spaces. Among the 47 approved in the Centelha Sergipe III preliminary Phase 2 result. |
| mandacaru | Intelligent document processing built in the STI/UFS context, with benchmarked model alternatives. |
| echo-womens-health-research-analytics | Survey-based evaluation of a remote women's health training programme. |
Every project states what it cannot show. The limits sections carry the assumptions that would change the conclusion, the data that is simulated and why, and the question the analysis could not answer. A result without its boundary is not a result.
Assumptions live in one file. Where a number is assumed rather than measured, it is named and isolated, so a reviewer can change it and re-run.
The decision comes first. Each README opens with the result and the decision it supports, then shows the evidence.
| Analytics | SQL · Python · pandas · DuckDB · dbt |
| ML and statistics | scikit-learn · LightGBM · PyMC-Marketing · MLflow · SciPy · statsmodels |
| Engineering | FastAPI · Docker · GitHub Actions · Plotly Dash · Streamlit |
| Marketing | GA4 · Meta Ads · Google Ads |
Trabalho na junção entre marketing e receita: mensuração confiável, análise que sobrevive a questionamento e modelos que terminam numa decisão (CAC, LTV, payback, margem de contribuição, retenção e incrementalidade).
São cerca de cinco anos entre mídia paga, growth em SaaS e analytics. O diferencial é construir a ferramenta em vez de parar no relatório. Os projetos acima têm código rodando, testes e premissas documentadas, organizados em um único método: validar o sinal, escalar o que funciona e reter o valor criado.
Cada projeto declara o que não consegue mostrar. É de propósito: premissa escondida é a forma mais cara de errar uma decisão de investimento.
ABADE · Strategy · Data · Growth