Ahad

“No labeled data, no reference noise channel, no prior knowledge of the interference frequency required.”

Two mechanisms work together. The passive architecture — LIF membrane filtering, Poisson rate encoding, population decoding — keeps clinical R-peaks intact at every noise level. STDP learning adds amplitude-domain SNR at low input SNR and pays for it in landmark fidelity. Phase 3 ran the whole thing on the MIT-BIH Arrhythmia Database and pinned down exactly where it wins.

Python · Numba @njit · uv · MIT-BIH · SQLite sweeps

“A finite automaton is a small printed object.”

Load one of eight presets or draw your own machine, then watch a vermillion token trace δ* across the diagram one symbol at a time. Diagnostics flag nondeterminism and unreachable states as you edit, undo is throttled so a long drag is one entry, and every machine fits in a URL.

Vite · React · TypeScript · Zustand + zundo · elkjs

“The only data that cannot be leaked is data that was never collected in the first place.”

Phase boundaries come from your own history, not a 28-day default. The home ring draws every phase twice, split at today — solid behind you, a hairline ahead — so the drawing itself is the marker. Logged days are filled, predicted days are hollow: a guess never looks like a fact.

Kotlin 2.2 · Compose · Room · Hilt · kotlinx-datetime

“Algorithms have personalities, and those personalities can be drawn.”

A recursive backtracker carves the maze on a pointy-top axial grid. Every solver is a generator that yields step descriptors, so one loop drives three algorithms it never has to understand. Each ink also has its own fill pattern, so the race stays readable under colour-blindness.

ES modules · Canvas 2D · rough.js · Web Animations · mulberry32