Pantas

1. Editorial "Ink & Rule" Design System & Source Sans 3

I stripped away the generic AI aesthetic (Poppins bold, stock blue #1F6BFF, drop shadows, emoji icons, pastel pills) in favor of an editorial printed-workbook identity. I used warm paper surfaces (#FBF9F4), near-black ink (#26221B), 1px hairline rules (#E5DFD3), and tabular figures, budgeting Board Green (#1D5C50) strictly for the single primary CTA per screen. I originally tried pairing a serif with sans, but on a phone screen the serif became an eyesore at small sizes and distracted from the content. I standardized on Source Sans 3 across all 15 typography roles to keep the focus entirely on reading.

Trade-off:Abandoning standard Material cards and elevated shadows meant writing custom layout math with hairline dividers, but it eliminated visual fatigue during 2-hour study drills.

2. Local-First Encrypted Persistence via Drift SQLite & SQLCipher

I designed the app under one core rule: studying is never blocked by the network; only account and money are. All question banks, user response logs, and scheduled drill states live in an encrypted local database using Drift SQLite with 256-bit AES SQLCipher for RA 10173 compliance. I replaced dynamic runtime CMS queries with an immutable, pre-compiled static SQLite seed (content.db / assets/seed/v1.json), ensuring cold boots and drill queries stay instant.

Trade-off:Bundling static seed databases adds ~8MB to the initial APK size, but guarantees zero-latency drills with 100% offline availability.

// Pure local database setup with SQLCipher encryption and static seed loader
LazyDatabase openConnection() {
  return LazyDatabase(() async {
    final dbFolder = await getApplicationDocumentsDirectory();
    final file = File(p.join(dbFolder.path, 'pantas_encrypted.db'));
    return SqlcipherDatabase(
      file,
      password: await secureStorage.getDatabaseKey(),
      setup: (rawDb) => rawDb.execute('PRAGMA cipher_memory_security = OFF;'),
    );
  });
}

3. Pure Dart FSRS-6 Spaced Repetition Engine & On-Device Optimizer

I implemented the FSRS-6 algorithm locally in pure Dart using the 21-parameter weight vector with dedicated same-day stability formulas for exam cramming behavior. To personalize intervals without sending study logs to a cloud server, I designed an on-device optimizer that fits parameters directly on device once a student logs 200 reviews (compared to Anki's 400+ threshold), guarded by a held-out test split to prevent overfitting.

Trade-off:FSRS-6's 200-review optimizer threshold is calibrated for short 8-week Philippine exam countdowns, using held-out split validation to reject overfitted weights while allowing students to revert to defaults in Settings.

// Pure Dart FSRS-6 scheduler with 21-parameter weights and same-day stability
FsrsItem scheduleFsrs6Review(FsrsItem item, Rating rating, DateTime now, List<double> w) {
  final elapsedDays = item.lastReviewed == null ? 0.0 : now.difference(item.lastReviewed!).inHours / 24.0;
  final nextStability = elapsedDays < 1.0
      ? fsrs6.calculateSameDayStability(item.stability, rating, w)
      : fsrs6.calculateStability(item.stability, item.difficulty, elapsedDays, rating, w);
  final nextDifficulty = fsrs6.calculateDifficulty(item.difficulty, rating, w);
  final intervalDays = fsrs6.nextInterval(nextStability, targetRetention: 0.90, decay: -w[20]);
  return item.copyWith(
    stability: nextStability,
    difficulty: nextDifficulty,
    due: now.add(Duration(days: intervalDays.clamp(1, 36500).round())),
    lastReviewed: now,
  );
}

4. Assessment Hub & Distractor Misconception Explanations

I replaced generic topic queues with an Assessment Hub (Today's Session, Drill, Retake, Mock Exam). Instead of showing a simple green checkmark or red cross, I authored answer reveals that explicitly explain why each incorrect choice (distractor) is wrong. To recreate real exam pressure, I wrote a custom canvas OMR bubble sheet with strict section boundary timers, question jump grids, and blueprint-weighted subject ratios.

Trade-off:Authoring custom misconception explanations for all 4 multiple-choice options quadrupled content writing time, but prevented students from relying on rote memorization.

5. Psychometric Integrity ("Never Invent a Figure")

I instituted a strict rule: never invent a figure. I banned synthetic score predictions and fake readiness percentages ('Passing Probability: 92%'). Instead, I structured the Progress tab around four honest questions: The Diagnosis (where points leak and why), The Mirror (behavioral archetypes like stamina drop-off and pacing under pressure), The Record (measurable movement over time), and What's Fading (FSRS decay curves). I show score impact strictly as point deltas ('Fixing these weak topics is worth +9 points').