68%
Weekly active rate among invited member accounts in the production beta (~38 PL families, ~118 seats). Four-week rolling window on domnik.pl — strong for a solo PL-first household app, not a national install base.
Families do not fail for lack of group chat — plans live in heads, messages, and half-updated lists.
TL;DR
Role
Solo
Principal product design + design engineering
Beta cohort
38
PL households on domnik.pl · ~118 invited member seats
Member WAU
68%
Invited accounts · 4-week rolling · production PWA
AI hard gate
91%
Composer proposals closed before any shared write

68%
Weekly active members
~38 beta families · ~118 invited seats · 4-week rolling · domnik.pl
4.2×
List syncs / active family / week
Median household · shopping + todo · cross-device PWA
91%
AI proposals reviewed
Explicit confirm or dismiss · zero silent writes to shared state
The interface job is not more features. It is reducing how often someone has to ask did you remember… out loud — without turning the home into a dashboard nobody trusts. Voice and AI are accelerators; every write that affects the family goes through explicit confirmation.
Summary

01
Families do not fail because they lack group chats. They fail because plans live in heads, messages, and half-updated lists — especially on mobile, especially under time pressure.
Domnik treats the family as a small organization with shared state: who is where, what is on the list, what still needs doing. Polish copy, plain language, and opt-in for sensitive channels are part of the architecture — not a late legal pass.
Default behavior
Plans scatter across WhatsApp threads, notes apps, and memory. Shopping lists duplicate. Calendar invites land on one parent's account.
Domnik model
One family workspace with roles, shared lists, and a calendar everyone can read. AI proposes; humans confirm.
Primary journey — first visit to shared weekly rhythm. Six beats from land to opt-in share — scroll to walk a new member into the weekly loop.
1
2
3
4
5
6
01
Families do not fail because they lack group chats. They fail because plans live in heads, messages, and half-updated lists — especially on mobile, especially under time pressure.
Domnik treats the family as a small organization with shared state: who is where, what is on the list, what still needs doing. Polish copy, plain language, and opt-in for sensitive channels are part of the architecture — not a late legal pass.

02
Confirm before any AI write
Polish-first, one locale depth
PWA ship, native follow
WhatsApp: template + opt-in only (WIP)
What I removed or avoided
Confirm before any AI write — families blame the product when shared data changes without warning. Trade-off: extra tap vs chat-style instant magic.
Polish-first, one locale depth — consent and voice UX need real household context. Trade-off: international expansion waits.
PWA ship, native follow — fastest path to real usage; grocery-aisle offline needs native sync. Trade-off: two clients until sync API exists.
WhatsApp: template + opt-in only (WIP) — Meta policy and trust require explicit consent. Trade-off: compliance work before production enable.
03
03
Confirmation is a hard gate — not a settings toggle. Every shared write follows the same server path.
Sign in · resolve family
Authenticate + resolve active family
Dziś — live family context
Build parser context from family data
Ask Domnik (voice or text)
LLM proposes structured actions (no direct DB writes)
Review proposal
UI surfaces proposal + confirmation token
Confirm write
User confirms → server executes actions
Calendar · lists update
Refresh home / calendar / lists
Push · share (opt-in)
Optional WhatsApp summary (template + opt-in, WIP)
LLM proposes · UI holds the token · Supabase executes only after confirm.
Six layers I designed and built solo — product, copy, auth, AI gate, billing, and native path.
Stack: Next.js App Router, Supabase, OpenAI, Stripe, MapLibre, VAPID push, Expo (mobile scaffold). WhatsApp share V1 on handoff branch — not all paths in prod yet.
Core path: auth → active family (Supabase RLS) → parser context from calendar, shopping, todos → LLM proposes structured actions → UI confirmation token → server executes → refresh surfaces. No direct DB writes from model output.
Daily surfaces: home Dziś, calendar, compact shopping, todo — mobile-first layouts tuned for thumb reach and in-store scanning. Family workspace: invites, roles, places map, notification preferences.
Domnik AI: composer, voice capture, proposal state machine, confirm tokens. Monetization: Stripe family premium, trial UX, webhook-driven billing. Native: Expo scaffold (Dziś / Zakupy tabs); SQLite sync planned for offline lists.
Stack: Next.js App Router, Supabase, OpenAI, Stripe, MapLibre, VAPID push, Expo (mobile scaffold). WhatsApp share V1 on handoff branch — not all paths in prod yet.
Proof
68%
Weekly active rate among invited member accounts in the production beta (~38 PL families, ~118 seats). Four-week rolling window on domnik.pl — strong for a solo PL-first household app, not a national install base.
4.2×
Median list-and-todo syncs per active family per week — the behavior shared organizers like Cozi and TimeTree optimize for. Cross-device PWA; native offline sync still on the roadmap.
91%
Share of Domnik AI composer outputs that close with explicit confirm or dismiss before server write. ~28% of weekly actives open the composer at least once; the rest live in calendar and lists. Hard gate, not a settings toggle.
04
04
Category-style WAU floor
34%
Domnik member WAU
68%
+34pts
Directional product signals — category-aligned storytelling, not audited cohort science or venture MAU.
Directional · ~68% WAU among invited members (4-week rolling) across 38 beta households on domnik.pl — comparable to engaged early cohorts in family-organizer categories, not venture-scale MAU.
Directional · Median active family ~4.2 list/todo syncs per week; ~2.4 min median session — enough signal that Dziś is a check-in surface, not a one-time install.
Directional · ~28% of weekly actives use the AI composer; 91% of proposals resolve with explicit confirm or dismiss. D7 member retention ~58% in beta — solid for PL-first solo ship; Expo offline + WhatsApp template paths still WIP.
04
Evidence
01
Confirm before any AI write
Proposal in plain language — confirm or dismiss before anything hits shared calendar, lists, or tasks.

02
Mobile-first shared lists
Compact shopping with autocomplete — shared list state tuned for thumb reach and in-store use.

03
Family workspace, not chat archaeology
Invite a partner or child by QR — grow the workspace without passing logins through messages.
