Health
NudgeMe
Natural-language reminders, designed to actually reach you
By Hiram BarskyFounder & Principal Designer
Role
Product Design · Full-stack · AI
Timeframe
2025
Published
July 2026
Outcome
1 typed sentence in → 5 delivery channels out
Built with
Figma Design · Claude Code — built end to end; confidence-tiered parsing · Next.js, Supabase, PWA · Netlify
The result
Live: type one plain sentence — 'call mom tomorrow at 5pm' — and NudgeMe schedules it, then actually reaches you across five channels: in-app, email, a calendar mirror, and true background push even when the app is fully closed.
Overview
NudgeMe turns a sentence like "call mom tomorrow at 5pm" straight into a scheduled reminder — no form, no date picker, no setup screen. The core promise, and the design constraint everything else answers to: type it once, NudgeMe remembers.
The problem
Reminder apps fail two ways, not one. Setup friction kills adoption before a habit even forms — every extra field is a reason to quit. And even for the reminders people do set, delivery is unreliable: most apps only notify while the app itself is open, so a due reminder in a closed tab just quietly never arrives. Solving only one of those problems isn't enough.
Research
Reminder apps don't lose people at one point — they lose them at two.
Setup friction
Every extra field is a reason to quit. Habits die in the gap between intending to set a reminder and finishing the form.
→ Drove: One typed sentence as the entire interface — parsing over forms, zero setup screens.
Silent delivery failure
Most web apps only notify while they're open. A due reminder in a closed tab just never arrives — and the user blames themselves.
→ Drove: Layered delivery: in-app, email, calendar mirror, and background push, each shipping and failing independently.
Wrong guesses destroy trust
One reminder scheduled for the wrong day teaches a user to double-check everything — which defeats the product.
→ Drove: Confidence tiers: act when certain, ask exactly one question when unsure, never silently guess.
The approach
- 1
Designed a confidence-tiered AI behavior model instead of a rigid form. High-confidence input creates the reminder with zero interruption. Medium confidence asks exactly one follow-up question. Genuinely unclear input asks the user to rewrite rather than silently guessing wrong and scheduling the wrong thing.
- 2
Treated delivery as a layered system to de-risk it, not one perfect channel to get right on the first try. Foreground in-app alerts shipped first, then email, then a Google Calendar mirror, then true background push once the app is fully closed. Each channel ships and degrades independently, so a failure in one never silently loses a reminder.
- 3
Designed the freemium model around a trust constraint, not just a pricing one. Free-tier limits apply only to new usage going forward. A feature a user already relies on for free never gets retroactively cut off just because a paid tier now exists.
Our thought process
The whole app answers to one sentence: type it once, NudgeMe remembers. Every decision — parsing, confidence tiers, delivery layers, even the pricing model — got tested against whether it kept or broke that promise.
What didn't work
The first instinct was a smarter form — fewer fields, nicer pickers. It was the same old friction with better paint, so it went in the bin and parsing became the whole front door. Treating delivery as one perfect channel failed the same way: a notification that only fires while the app is open isn't a reminder, it's a coincidence. That failure forced the layered system.
The outcome
A reminder that reaches you whether the tab is open, the app is fully closed, or you're just looking at your calendar — and a monetization model that never breaks a promise it already made a user.
What this means for your project
Have an AI feature that has to act on messy human input? This is the pattern we'd bring: behavior tuned to confidence — act when it's sure, ask exactly one question when it isn't, never guess wrong — paired with a system that degrades gracefully instead of failing silently.
“One of the most talented designers I have had the pleasure to work with.”
