LAB//INDEX

The Lab

The personal laboratory, 2021–today: experiments nobody commissioned. Some became systems, some remain questions. All of them are real.

LAB/01 — SIGNALS

Decoding "A Sign In Space"

In 2023 ESA transmitted a simulated extraterrestrial message from the ExoMars TGO probe, challenging anyone to decode it. My attack: bit-plane visualization of the 256×256 message and an unconventional hypothesis — the message as a program for a single-instruction FlipJump machine, implemented and run on the real data, with the memory evolution animated.

The FlipJump machine memory evolving on the real message data
The FlipJump machine memory evolving on the real message data
LAB/02 — CRYPTOGRAPHY

A neural network vs RSA

Can a neural network learn to factor semiprimes? A serious experiment on an impertinent question: multi-input architecture (product bits + square root), outputs over both prime factors, and a custom loss weighted by bit significance. (Spoiler: RSA is safe — but the why is instructive.)

LAB/03 — QUANTA

Animated quantum mechanics

A complete sub-lab: time-independent Schrödinger solvers, the Numerov method, Hartree methods for the structure of matter, and the time evolution of interacting identical particles rendered as videos.

Interacting identical particles — joint probability density evolving in time
LAB/04 — DYNAMICS

Rigid bodies from first principles

A 6DOF simulator written from scratch: Euler equations in the body frame, quaternion attitude integration, constraints implemented from a published derivation — validated on free-fall, pendulum and unstable-axis spin test cases.

LAB/05 — COMPLEXITY

Economics emerging from agents

Agent-based models: wealth inequality emerging from simple exchanges, seller-shop markets, and early experiments with orchestrated LLM agents — dated March 2024, before it became fashionable.

Wealth distribution emerging from simple random exchanges
LAB/06 — LANGUAGE

A GPT from scratch, in Italian

A GPT-style LLM trained from scratch on an Italian corpus: custom BPE tokenizer, hand-written PyTorch training loop, optimization experiments. Understanding transformers by building one.

Many of these experiments live in private repositories; the code is gladly shown on request.