惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

V
V2EX
J
Java Code Geeks
月光博客
月光博客
博客园_首页
The GitHub Blog
The GitHub Blog
Vercel News
Vercel News
B
Blog RSS Feed
博客园 - 聂微东
宝玉的分享
宝玉的分享
T
Tailwind CSS Blog
Jina AI
Jina AI
S
SegmentFault 最新的问题
B
Blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
有赞技术团队
有赞技术团队
Hugging Face - Blog
Hugging Face - Blog
Google DeepMind News
Google DeepMind News
阮一峰的网络日志
阮一峰的网络日志
The Cloudflare Blog
量子位
Martin Fowler
Martin Fowler
博客园 - Franky
大猫的无限游戏
大猫的无限游戏
博客园 - 叶小钗

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
The most important feature in our AI journaling app is th...
Vineet Negi · 2026-04-25 · via DEV Community

Vineet Negi

Day 7 of building Evengood — a 60-second end-of-day reflection app — in public on the Build with MeDo hackathon.

Today I shipped the feature I'm most proud of in seven days of building.

It's the feature where the AI doesn't do anything.

The problem

Every AI journaling app I've ever tried has the same failure mode. You type something heavy — "my dad died last week," "I had a panic attack on the train," "I think I'm getting fired Monday" — and the LLM cheerfully reframes it into a growth-mindset Hallmark card. "What a meaningful day of self-reflection! Here are three things to be grateful for..."

It's the worst possible response. It tells the user the product wasn't built by anyone who has ever had a hard day.

And it's the default behavior of every "wrap GPT in a textarea" app shipped this year.

What I shipped today

Quiet Mode. A pure-function detector that runs on the user's reflection text before it ever leaves the device. If the text contains signals from four categories — grief (died, funeral, passed away), crisis (panic attack, breakdown, can't breathe, want to die), breakup/loss (broke up, divorce, fired, miscarriage, diagnosis), or extreme overwhelm (I can't do this anymore, I give up) — the app does less, not more.

Specifically:

  • The Gemini reframe call is skipped entirely. No API hit, no token spend, no AI rewriting your grief into a productivity tip.
  • The text-to-speech call is skipped too.
  • Instead, a soft acknowledgment card replaces the reframe card:

Some days don't need reframing.

Be gentle with yourself tonight. This one's saved, untouched.

Quiet Mode · we noticed this might be a heavy day

  • The entry is saved to history with isQuiet: true. Original text preserved exactly. No reframe text stored.
  • The weekly Pattern detection (also a Gemini call, runs over the last 7 entries) excludes quiet entries from its corpus. The pattern is about your rhythms, not your tragedies.
  • Streak still counts. Quiet days are still days you showed up.
  • For self-harm signals only, one calm line appears below the card: "If you need to talk to someone right now: text or call 988 (US) or befrienders.org (global)." No emoji, no red, no popup, no urgency styling. One line.

The whole thing is invisible until it's needed. There's no banner saying "Quiet Mode protects you," no settings tour, no onboarding card. It just sits there silently and does nothing 99% of the time. The 1% it does something, it does the smallest possible thing.

Why this is the most important feature

Hackathon judges in 2026 have watched a thousand "we wrapped an LLM" demos this season. The thing that cuts through is a product that demonstrates judgment about when not to use the LLM. Quiet Mode is that demonstration in a single 15-second beat:

  1. Type "my dad died last week"
  2. Hit share
  3. App goes quieter, not louder

That's the demo. That's the whole pitch.

It's also the thing that makes the product feel like it was made by humans who have thought about humans.

How it's built

The detector is a 40-line pure function in src/lib/quietMode.ts:

export function detectQuietMode(text: string): { quiet: boolean; signal?: string } {
  if (!text) return { quiet: false };
  const lower = text.toLowerCase();
  for (const signal of QUIET_SIGNALS) {
    const re = new RegExp(`\\b${signal.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}\\b`, 'i');
    if (re.test(lower)) return { quiet: true, signal };
  }
  return { quiet: false };
}

Enter fullscreen mode Exit fullscreen mode

No LLM. No external call. Word-boundary regex against a hand-curated signal list. Runs in microseconds. Costs zero. Privacy-respecting (the matched signal is stored only in the local entry, never logged or sent anywhere).

The submit handler in App.tsx checks the detector before deciding what to do:

const { quiet, signal } = detectQuietMode(reflectionText);
if (quiet && quietModeEnabled) {
  saveQuietEntry({ text: reflectionText, signal, date: today });
  showQuietAcknowledgment(signal);
  return; // no Gemini call
}
// otherwise: normal reframe flow
const reframe = await callGemini(reflectionText);

Enter fullscreen mode Exit fullscreen mode

The Pattern detection edge function (Supabase) added a single filter:

const entries = allEntries.filter(e => !e.isQuiet);

Enter fullscreen mode Exit fullscreen mode

That's the whole feature. ~80 lines of code. Half a day of work. The hardest part was deciding what not to add.

What I deliberately didn't build

  • No popup or modal. Quiet Mode never interrupts you to congratulate itself.
  • No emoji on the acknowledgment card. This isn't a moment for sparkles.
  • No red or alarming colors. The card uses the same calm lavender→peach gradient as the rest of the app.
  • No "we detected crisis keywords" language. The user-facing copy says "we noticed this might be a heavy day." One word change, completely different feeling.
  • No analytics event when Quiet Mode triggers. We don't need to know.
  • No required settings tour. It's on by default and you can find the toggle if you want to turn it off.
  • No fancy ML classifier. A regex list curated by hand is more honest about what we're doing — looking for specific words — than a black-box model that might silently change behavior.

The restraint is the feature.

What's live right now

Evengood, Day 7 of 30:

  • 60-second voice or text reflection (whisper for transcription)
  • Calm AI reframe via Gemini (when not in Quiet Mode)
  • Weekly Pattern detection — your last 7 reflections turn into one sentence
  • Moment of the Week — one keepsake auto-pulled from the week
  • Save-as-image keepsake (lavender/peach card with your moment)
  • Streak counter, history view, ambient sound (rain / piano / silence)
  • Email opt-in for the Sunday Pattern
  • Quiet Mode (today's ship)

No account. No password. Nothing about you leaves the device except an anonymous device_id and (if you opt in) your email for the Sunday note.

Try it

If you build AI products: I think the next year of differentiation isn't going to be about which model you wrapped. It's going to be about whether you have the judgment to not call the model.

More tomorrow.

— Day 7 / 30 🌿

BuildInPublic #BuildWithMeDo