03

// domain 03

Neurish

Neurish and Friends With Epilepsy: platforms engineered for people in vulnerable moments. Trust before features, intention before engagement, reliability when nobody is watching.

NuxtPostgresTimescaleDBWebSocketsAnthropicOpenAIVercel

// why neurish exists

A calm place for a hard condition.

Neurish is a behavioral platform for the epilepsy community. Daily reflection, pattern recognition, gentle community. The mission is small enough to mean something: be the calmest health-adjacent app a person installs that year.

Most apps for chronic conditions optimize the wrong loop. They try to maximize check-ins and notification opens. Neurish optimizes for the opposite: minutes spent, not sessions started; insight delivered, not attention extracted. Intentional community over algorithmic feed.

// friends with epilepsy

A community, not a feed.

FWE is the social companion to Neurish. The platform's goal is meaningful interaction, not metrics that look like meaning.

The design language assumes people are tired, recently post-seizure, or quietly worried about something they have not said aloud. The interaction model assumes the same. Posts are short by default; replies are slower than X / Twitter on purpose; the home page never displays an unread count.

// behavioral design

Onboarding and the loops that follow.

The first ten minutes of any platform are an honest promise about the next ten thousand. I design them slowly.

// 01 onboarding

Nine questions, not three.

The longer onboarding outperforms the short one because it asks the right things, not because length wins.

// 02 activation

First useful moment.

Defined per-user, not per-app. The activation event for a caregiver is different from the one for a person with epilepsy.

// 03 engagement loops

Optional by design.

Every loop has an opt-out at the same depth as the loop itself. No coercive defaults.

// 04 response velocity

Slow is sometimes the feature.

Replies surface in waves rather than instantly. Quality goes up; pile-on goes down.

// notification systems

Wave notifications.

Instead of a single push when something happens, the system queues events into thoughtful waves: a morning summary, an afternoon nudge, an evening reflection. Behavioral targeting chooses which wave a given person belongs to.

The anti-spam philosophy is the opposite of the industry default: assume each user has a budget of attention per day, and underspend it.

neurish.svg · wave · queue

topology

// trust & moderation

An architecture, not a policy.

Moderation outcomes are downstream of system design. Get the design right and policy stays small.

  1. 01

    Reputation accrues slowly

    New accounts cannot do harmful things, even legitimately. The system is patient because the people are.

  2. 02

    Moderation is part of the product

    Not bolted on. Mod tooling shares its design language with the main app, and moderators are treated as a first-class user type.

  3. 03

    Anti-spam is anti-noise

    A loud account is treated like a malfunctioning sensor: quieted, then investigated.

  4. 04

    Emotional safety is observable

    Tone, escalation, and pile-on patterns are surfaced as metrics, not just feelings.

// metrics & signals

What I watch, what I refuse to optimize.

Engagement metrics are easy and dangerous. These are the ones that survived a year of being interrogated.

wave open rate

63%

vs 14% for batch notif

d7 retention

46%

fwe cohort · q1 2026

response p50

4h 12m

time-to-helpful-reply

spam reports

0.7 / kU

per 1k user-days

// values are illustrative; production numbers live behind an auth wall.

// architecture diagrams

Onboarding, engagement, behavioral engine.

The behavioral engine sits between the app and the data store, deciding which wave a user belongs to and which surfaces deserve which message.

behavioral-engine.svg

topology