Case 01

Praça Araguaia

Context

In southern Pará, ranchers decide when to sell based on prices that arrive by WhatsApp and word of mouth. The information exists and is public, but it is scattered across sources that do not speak the language of the people working the land — and it arrives too late for anyone closing a deal today.

What I built

Twelve price feeds gathered by a scheduled job, stored in Postgres and published with revalidation — plus a daily bulletin, rainfall forecast and the market thermometer, which turns the whole set into a single reading: is the market hot or flat.

Case 02

Capital Financeiro

Context

Fifteen years doing the same reconciliation by hand: turn PDF into rows, rows into categories, categories into a decision. The work never changes, only the month — and a typo only surfaces when the total refuses to add up, if anyone checks.

What I built

The PDF is parsed inside the browser and never leaves the machine. The app extracts the entries and checks their sum against the total printed on the document itself before storing anything; only then does it categorise, split by billing period and draw the charts.

Case 03

Vaga Certa

Context

Job hunting means reading dozens of listings to find out almost none fit. The filtering is manual, repetitive and exhausting long before it pays off — and career changers are exactly the people least able to judge on their own whether their profile fits.

What I built

The AI reads the résumé and extracts role, seniority and skills; three public sources are queried in parallel and every opening comes back with a score, the points in favour and what the profile is missing. The model output is validated against a schema before it reaches the screen — no sign-up, no résumé stored.

Case 04

gabarito_AI

Context

An exam notice runs forty pages and turns into months of study — and almost every candidate does that translation by hand, in a spreadsheet that goes stale within a week. The heavy lifting happens before any studying starts, and that is where most people give up.

What I built

The AI reads the whole document — PDF, photo or scanned, through the OCR of the model itself — builds the plan by subject, weighted the way the board actually tests it, and transcribes the exam questions in independent batches, so one failed batch does not take the rest down. Every record is isolated per user in Postgres.

Case 05

Rendimento

Context

Every bank simulator shows the gross figure, and gross is the one number nobody ever receives. After the regressive tax and the inflation of the period, what is left is something else — and comparing savings, CDB and Treasury bonds means putting all three under the same rule, which a spreadsheet rarely does.

What I built

A pure calculation engine, covered by tests, projects the future value or solves for the contribution a goal requires, always in present-day money. Eleven investments enter the same scenario and are ranked by net return; the app runs entirely in the browser, with no backend and no data collection.