<h2>The Problem</h2>
<p>Everyone wants to journal. Almost nobody does.</p>
<p>The barrier isn't motivation — it's friction. After a full day of work, the last thing anyone wants to do is open a blank page and write about it. Traditional journaling apps like Day One or Notion still require you to sit down and type. That's the problem: <strong>journaling demands effort at the exact moment you have the least of it.</strong></p>
<p>Meanwhile, you already track what you do. You check off tasks, take photos, send voice notes. That data tells the story of your day — it just disappears after you're done with it.</p>
<h2>What I Built</h2>
<p><a href="https://isip.pastelero.ph">Isip</a> (Tagalog for "mind/thought") is an iOS app that turns the things you already do — completing tasks, speaking thoughts, snapping photos — into journal entries automatically. You never open a blank page. You never write a single word. The journal writes itself.</p>
<h3>Screenshots</h3>
<p><img src="/assets/project/isip/01_hero.png" alt="Voice input — say it and AI classifies it"></p>
<p><img src="/assets/project/isip/02_tasks.png" alt="Tasks become journal entries automatically"></p>
<p><img src="/assets/project/isip/03_journal.png" alt="Daily journal log with summaries and tags"></p>
<p><img src="/assets/project/isip/04_insights.png" alt="Insights — mood trends, streaks, stats"></p>
<p><img src="/assets/project/isip/06_watch.png" alt="Apple Watch companion — capture from your wrist"></p>
<p><img src="/assets/project/isip/05_onboarding.png" alt="100% offline, no account, private by default"></p>
<h2>How It Works</h2>
<ol>
<li><strong>Capture anything</strong> — Type a task, speak a thought, or take a photo. Isip's AI classifies it automatically (task vs. journal entry) and sets reminders if it detects a time.</li>
<li><strong>Check off tasks</strong> — Complete tasks throughout the day. Each one feeds into your journal.</li>
<li><strong>Journal appears</strong> — AI weaves your completed tasks, voice notes, and photos into a cohesive daily journal entry with mood detection, tags, and a reflection prompt.</li>
</ol>
<h3>Core Features</h3>
<ul>
<li><strong>Auto-journal from tasks</strong> — Complete a task and AI generates a warm, second-person journal entry from it</li>
<li><strong>Voice capture</strong> — On-device speech-to-text transcription for hands-free input</li>
<li><strong>Photo analysis</strong> — Vision framework analyzes photos (scene classification + OCR) and generates journal insights</li>
<li><strong>Smart reminders</strong> — Say "meeting at 12:30 PM" and AI creates the task and sets the reminder</li>
<li><strong>Smart classification</strong> — AI auto-classifies input as task or journal entry with confidence scoring</li>
<li><strong>Conversational journaling</strong> — Interactive AI chat that asks reflective questions and synthesizes responses into a journal entry</li>
<li><strong>Mood tracking</strong> — Five-point emoji scale with historical trends</li>
<li><strong>Insights dashboard</strong> — Streaks, mood charts, top tags, "On This Day" flashbacks</li>
<li><strong>Apple Watch companion</strong> — Voice-first interface for capturing thoughts from your wrist</li>
<li><strong>Markdown export</strong> — Full data portability, no lock-in</li>
</ul>
<h3>The Privacy Promise</h3>
<p>Everything runs on your device. No account. No cloud. No server. No sync.</p>
<ul>
<li>All AI processing uses Apple's Foundation Models framework — on-device, never sent anywhere</li>
<li>Speech recognition uses on-device mode</li>
<li>Photos are analyzed in memory, never uploaded</li>
<li>Data stored locally with <code>NSFileProtectionComplete</code> encryption</li>
<li>Delete the app and the data is gone — because we never had it</li>
</ul>
<h2>Tech Stack</h2>
<table>
<thead>
<tr>
<th>Layer</th>
<th>Technology</th>
<th>Why I Chose It</th>
</tr>
</thead>
<tbody>
<tr>
<td>Framework</td>
<td>SwiftUI + Swift 6</td>
<td>Native iOS with modern concurrency</td>
</tr>
<tr>
<td>AI</td>
<td>Apple Foundation Models</td>
<td>On-device inference, structured output via <code>@Generable</code></td>
</tr>
<tr>
<td>Persistence</td>
<td>SwiftData</td>
<td>Local SQLite with <code>@Observable</code> and <code>@Query</code></td>
</tr>
<tr>
<td>Vision</td>
<td>Vision framework</td>
<td>On-device scene classification and OCR</td>
</tr>
<tr>
<td>Speech</td>
<td>Speech framework</td>
<td>On-device speech-to-text transcription</td>
</tr>
<tr>
<td>Watch</td>
<td>WatchConnectivity</td>
<td>Bidirectional iPhone ↔ Watch communication</td>
</tr>
<tr>
<td>Build</td>
<td>XcodeGen</td>
<td>Reproducible project generation from <code>project.yml</code></td>
</tr>
<tr>
<td>Testing</td>
<td>Swift Testing</td>
<td>Modern test framework with <code>@Test</code> macros</td>
</tr>
</tbody>
</table>
<h2>Architecture</h2>
<h3>AI Service — 8 Distinct AI Functions</h3>
<p>The <code>AIService</code> wraps Apple's Foundation Models with typed, structured output:</p>
<pre><code>import FoundationModels

@Observable
@MainActor
final class AIService {
    private var session: LanguageModelSession?
    var availability: SystemLanguageModel.Availability
}
</code></pre>
<p>Each AI function uses <code>@Generable</code> types for predictable structured responses:</p>
<pre><code>@Generable
struct JournalInsight: Sendable {
    @Guide(description: "A concise daily summary in 2-3 sentences")
    var summary: String

    @Guide(description: "Key themes as single-word tags", .maximumCount(5))
    var themes: [String]

    @Guide(description: "A thoughtful reflection question")
    var reflectionPrompt: String

    @Guide(description: "Detected mood", .anyOf(["great", "good", "okay", "low", "rough"]))
    var detectedMood: String
}
</code></pre>
<p>The 8 AI functions:</p>
<ol>
<li><strong>Task → Journal Insight</strong> — generates summary, themes, reflection prompt, mood</li>
<li><strong>Photo → Journal Insight</strong> — creates entry from Vision-analyzed photos</li>
<li><strong>Chat Journaling</strong> — generates reflective follow-up questions across 4 turns</li>
<li><strong>Journal Merge</strong> — intelligently weaves new content into existing entries</li>
<li><strong>Journal Analysis</strong> — extracts mood, themes, summary per entry</li>
<li><strong>Text Classification</strong> — task vs journal with confidence threshold</li>
<li><strong>Reminder Extraction</strong> — detects times in natural language and sets reminders</li>
<li><strong>Graceful degradation</strong> — full app functionality without AI</li>
</ol>
<h3>Capture Flow</h3>
<pre><code>User Input → CaptureService.submitText()
  → CaptureClassifier.classify() [AI]
    → isTask? → Create TodoItem + extract reminder time
    → isJournal? → Merge into today's JournalEntry [AI]
  → Update DaySummary
  → Refresh insights [AI]
</code></pre>
<h3>Watch Communication</h3>
<pre><code>Apple Watch: Voice → WatchConnectivityService.send()
  → transferUserInfo(payload) via WCSession
  → iPhone: PhoneConnectivityService.receive()
    → CaptureService.submitWatchCapture()
    → Merge into journal
</code></pre>
<p>If the iPhone app isn't active, captures are buffered in UserDefaults and drained when the app opens.</p>
<h3>Data Model</h3>
<ul>
<li><strong>TodoItem</strong> — tasks with priority, due date, completion tracking</li>
<li><strong>JournalEntry</strong> — one per day, with AI metadata (summary, tags, mood, reflection prompt)</li>
<li><strong>Tag</strong> — reusable theme tags with inverse relationships</li>
<li><strong>DaySummary</strong> — aggregated daily stats</li>
</ul>
<p>Architecture follows Apple's recommended MV pattern (not MVVM) with SwiftData <code>@Observable</code> classes and <code>@Query</code> for reactive data fetching.</p>
<h2>What Building This Taught Me</h2>
<h3>1. Apple Intelligence is surprisingly capable with structured output</h3>
<p>The <code>@Generable</code> protocol with <code>@Guide</code> annotations gives you structured, typed output from the language model — similar to OpenAI's JSON mode but running entirely on device. The quality is good enough for journal summaries, mood detection, and text classification without any cloud fallback.</p>
<p>The key is writing good system prompts. A warm, second-person tone in the system instruction produces journal entries that actually sound human.</p>
<h3>2. The merge problem is harder than generation</h3>
<p>Generating a journal entry from scratch is easy. The hard part is <em>merging</em> new content into an existing entry without losing context or creating repetition. When a user completes a task at 9 AM and another at 3 PM, the AI needs to weave both into one cohesive narrative.</p>
<p>I solved this by giving the merge prompt the full existing entry and instructing it to integrate the new content while preserving everything that's already there.</p>
<h3>3. Graceful degradation is non-negotiable for on-device AI</h3>
<p>Apple Intelligence requires specific hardware (iPhone 15 Pro+) and the model might not be ready immediately. The app needs to work completely without AI — basic task management and manual journaling. AI features are additive, not required.</p>
<h3>4. WatchConnectivity is simple but the edge cases aren't</h3>
<p><code>transferUserInfo</code> is reliable and queues messages when the phone is unreachable. But you still need a buffering layer on the iPhone side for cases where the app isn't active when the message arrives. UserDefaults with a drain-on-launch pattern handles this cleanly.</p>
<h2>What I'd Do Differently</h2>
<p>Start with the Apple Watch experience first. The voice-first, zero-UI capture on the Watch turned out to be the most compelling input method. If I had designed the capture flow around the Watch constraints from the beginning, the iPhone UX would have been even simpler.</p>
<h2>Links</h2>
<ul>
<li><strong>Website:</strong> <a href="https://isip.pastelero.ph">isip.pastelero.ph</a></li>
</ul>


Projects.

Isip: The Journal That Writes Itself