Pedro Mariano / portfolio
Lisbon · Business Management, ISCAL

Pedro
Mariano.

I build automation on real business data: systems that read, check and act, and that a person can always verify.

Scroll
Pedro Mariano

About

Business problems, solved with software that can be checked.

I study Business Management at ISCAL in Lisbon, finishing in 2027, and I run operations and systems at my family's minimarket. There I built a system that reads supplier invoices, proves them against the Portuguese fiscal QR code and proposes new prices for a person to approve.

What I care about is the part that makes automation safe: tests, audit trails, backups and a way to undo every change.

More about me, my experience and my CV →

Python · SQL · Power BIplus Azure, n8n and local AI models
Finance and taxAccounting (SNC), auditing, IRS and IRC
Portuguese · Romanian · Englishnative, native, C1

Projects

Things I built, ready to try.

01

Invoice → PriceLive

Reads supplier invoices, checks every line against the fiscal QR code VAT rate by VAT rate, and proposes new costs and shelf prices that a person approves. Photograph a real Portuguese invoice with your phone and watch it being read and proved, in the store app's own design. Nothing leaves your device.

  • Python
  • OCR
  • Fiscal QR
  • 22 tests
  • Safe writes with undo
02

StratForgeLive

A trading strategy builder: backtests with fees and slippage, tests that catch overfitting, live Bitcoin candles, and an AI analyst that reads the chart and calls the engine instead of guessing.

  • TypeScript
  • React
  • AI agent
03

Pricing health dashboard

A Power BI model that answers one question: where is the store losing margin because prices fell behind costs? Products sold below cost are ranked by money lost.

  • Power BI
  • DAX
  • Star schema
04

Invoice triage workflow

Each new invoice is summarised, logged to a spreadsheet and routed by email: "ready to approve" or "needs a person", with the reasons. It never writes a price.

  • n8n
  • Google Sheets
  • Gmail
05

Does an AI model predict Bitcoin?

I connected Kronos, an open model trained only on price candles, to StratForge and tested it on a year of Bitcoin data it had never seen. It showed no reliable edge. The write-up explains how I measured it.

  • PyTorch
  • GPU
  • Out-of-sample test

Contact

Let's talk.

I am looking for a hybrid role in Lisbon, or a remote one, in business and finance systems, automation or AI.