Start
One call and one question
Thirty minutes, and one question that decides the rest: which decision rests on this number? If the answer is "none", I say so, and there is no project.
Market intelligence and applied research
A decision that rests on a number needs the number and the path to it.
I do market intelligence and applied research for the people who will decide with that number, and I hand over how it was built alongside it.
Services
What I do.
I turn difficult sources and imperfect data into evidence you can use for a decision. The four fronts below are not separate areas: they are four routes to the same delivery, and a project usually passes through more than one.
-
01
Understand the market
Sizing · competition · entry
I size the opportunity, benchmark the competitors and read the media and market sources behind an entry decision. That is what I did for two years for Asian companies arriving in Brazil, building the Brazil and Portugal operation of a Hong Kong firm from zero.
-
02
Measure
Econometrics · panel · series · spatial
Econometrics, time series, panel data and spatial models, with an explicit identification design and correction for multiple testing. I deliver the code that reproduces the number, not only the number.
-
03
Build the data
PDF · XML · microdata · administrative bases
I turn unstructured documents (PDF financial statements, tax XML, administrative microdata) into a base with provenance, per-record status and arithmetic reconciliation. Where the read fails, the record is marked as failed, not as zero.
-
04
Operationalise
Classifiers · agents · auditable systems
Classifiers and agents with intent-based limits, human review and an audit trail. The model turns already-verified facts into a thesis, risks and questions. It does not choose on its own and it files nothing.
The work
How it works.
How it starts, what you receive, and what I say before accepting.
Scope
Written before any invoice
I send back in writing what is included, what is left out and what I cannot guarantee.
Delivery
The result with its path attached
You receive the number, the raw data, the code that reproduces it and the list of what was decided after looking at the data. Someone else can redo it.
Limits
What I do not do
I do not promise results that depend on a third party, and I do not call a sample I had already seen a validation.
The deadline goes in the scope, not here. It depends on the source, and a public source that only publishes through a form, with no stable URL, changes the arithmetic.
Evidence
What I built.
Not all of it is a product: some is published research, some is an internal tool. What they share is being written so an outsider can check them. The numbers below come from the repositories.
Benevente
A pre-registered multifactor study on B3, Brazil's exchange, built so the methodology can be audited and the result can fail. The price panel is rebuilt from COTAHIST, so delisted firms stay in the universe. The original panel was silently dropping 166 of 514 issuers.
- 73 trials audited with the deflated Sharpe ratio
- 8 hypotheses rejected, published in the same detail
- 7 methodology defects I found and fixed myself
Synthetic Reader Panel
Four fictional readers with sharply different criteria critique a text (a paper, a thesis, a valuation, a CV) and return a score, the specific passages that bothered each one, and a revision.
- 4 personas per text type, two fixed and two variable
Financial statement extractor
A published balance-sheet PDF goes in; one row per company comes out, with legal name, tax ID, fiscal year, scale, scope, value and status. Each record carries its own status: a read error comes out as an error, never as a zero.
- 15 real statements as the calibration corpus
- 4m53s on a 30-file batch, against 20 min before
Cidade do Concurso
Aggregators answer which civil-service exams are open. This one answers which post is worth the life you want, crossing a proprietary municipal quality-of-life index with salary adjusted for real local cost of living.
- 5,571 municipalities in the IBGE territorial base
Occupational exposure to AI
Which Brazilian municipalities are most exposed to generative AI, and can the measure available today identify them? Exposure comes from the ILO occupation scores. The hard part is each city occupational composition, which no single source observes in full.
- 5,570 municipalities geocoded
Santa Teresa microclimate
Climatology of a mountain valley in Espírito Santo from INMET and INCAPER series: extremes, which month rainfall actually concentrates in, whether the microclimate is drying out, and a rainfall predictor.
Every number above traces back to a repository or an audit artefact. If one does not reconcile, it is a bug, and I want to hear about it.
Who
Diego Gallina de Oliveira
I am an economics undergraduate at Fucape Business School, on a dual degree with Business Administration. Three years as a funded research assistant, CNPq at Ifes and then Fapes at Fucape, left me a habit: look first at how the sample was built and at what was chosen after seeing the data.
Since 2024 I have done market intelligence for Wisers, of Hong Kong, where I built the Brazil and Portugal operation from zero. I have presided over Fucape Jr. since 2025, re-elected, and opened DGO in May 2026 to do the same kind of work under my own contracts, in Portuguese, English and Spanish.
- 2021 to 2023CNPq / Ifes · funded research in physics, astronomy and technology
- 2023 to nowFucape Business School · Economics and Business Administration, expected 2027
- Mar to Nov 2024Fapes / Fucape · applied research, and the 78 municipalities assessed under the ITGP methodology
- Dec 2024 to nowWisers Information, Hong Kong · Brazil and Portugal operation built from zero
- Aug 2025 to Mar 2027Fucape Jr. · president, re-elected
- since May 2026DGO Foresight & Intel · market intelligence and applied research under my own contracts