Full catalog

Everything I build

Eleven systems across five fronts — from live products to AI agent safety and the apps my family uses every day.

Pillar 1

Live products

It's real — you can click it now.

AnnyStudy

Live

My daughter photographs her notes. The AI does the rest — and even discusses the subject with her.

A study platform that's live: a photo of the notebook becomes a spaced-repetition flashcard, generates quizzes, debates the content via the Feynman technique, has an ENEM mode and a built-in economy simulator. Born from a real problem at home.

CPA-10 web

Live

An AI that refuses to give you the answer — it argues with you until you understand.

A study app for Brazil's ANBIMA CPA-10 certification: each lesson requires discussing with the AI before the quiz, instead of rote memorization. I built it for my daughter Drika to pass the exam.

Pillar 2

Quantitative engineering

The robust private and the auditable public reinforcing each other.

Bandit Thompson

Private

My algorithmic trading system — years of research treated as a lab, not a hunch.

A bot that trades gold (XAUUSD) in production, built on the scientific method: every decision becomes a hypothesis, tested against market noise, and only stays if the data proves a real edge. Custom infrastructure to separate luck from edge, a notebook of documented findings, 100% deterministic. The proof it's serious: when the analysis showed the algorithm the bot is named after added nothing, I removed it and kept only what held up. Leaner, faster, stronger.

backtest-optimizer

Open source

The method behind my bots, open for anyone to audit.

Open-source: the protocol I use to optimize a strategy without falling into the trap of fitting the bot to a past that never repeats. It's the public face of the discipline behind Bandit Thompson.

v8-observer

Private

My bot trades real money. I built a second system whose only job is not to trust it.

A forensic observer that records, every 15 minutes, everything the trading bot does and the context it did it in — never able to interfere, by contract: read-only. When the data channel started rotting on a recurring basis, the observer began detecting the failure and recovering on its own. Over the two weeks measured: 77 autonomous recoveries, all successful, without a single human intervention.

EUR/USD Core B

Private

I tested six years of EUR/USD against its own noise; what survived became a bot — and every decision was stamped before I saw the result.

End-to-end quantitative research on EUR/USD: six years of data, out-of-sample walk-forward validation, and decisions locked by gates defined before seeing the result. The confirmed thesis became a bot, now under test on a cloud demo account. It's the Bandit Thompson method applied to a second instrument — proof the edge wasn't luck.

Pillar 3

Financial intelligence

Market study turning into a tool.

FII Analyzer

Private

I took my study of the financial market and turned it into an analyst that never sleeps.

Analyzes Brazilian paper REITs with an institutional buy-side stance: skeptical, decompositional, never buying the manager's narrative. A 9-dimension framework, every claim cited to its source, a categorical verdict at the end. It doesn't recommend — it investigates.

MacroSim

Private

I don't program inflation. I build the people, and inflation emerges on its own.

A simulator where households and firms make simple decisions and macro phenomena — inflation, unemployment, recession — emerge from collective behavior. Agent-based computational economics: the bridge between the financial market and code.

Pillar 4

AI agent safety

Where consent leaves the software and becomes hardware.

Governador

Private

I took the 'approve' button out of the AI agent's own hands and put it on a physical key it can't reach.

A physical out-of-band consent device for AI agents. When the agent requests a sensitive action, the request travels over Bluetooth to a handheld device, and the decision — approve or deny — happens on a physical key, outside the software the agent controls. Three pillars: out-of-band, least privilege, non-repudiation. MVP validated on real hardware and wired into Claude Code.

Pillar 5

Family apps (in use)

In use by my family, every day.

Família Landi

Private

Knowing the people I love got there safely shouldn't depend on a 'made it' text.

A real-time location app my family uses every day. Everyone shows up on the map, and the app notifies when someone arrives at or leaves a saved place — working with the phone locked and the app closed. Flutter + Firebase, running on the family's phones.

Anny Bridge

Private

If I've become the family's tech support, at least I automated the boring part.

A remote diagnostics bridge: an app on my wife's phone talks to a Telegram bot via n8n, and I inspect and help her device from afar — without asking for a screenshot at every problem.

The door

If you got this far, send a message.

Even if just out of curiosity. I reply on LinkedIn or by email.