Smart Suggestions¶

ShopSync uses machine learning to provide intelligent suggestions as you type, making item entry faster and more accurate.

Overview¶

Smart Suggestions leverage TensorFlow Lite to predict:

• Item names
• Appropriate icons
• Categories
• Common quantities

How It Works¶

On-Device ML¶

ShopSync uses an on-device ML model:

• Privacy-focused - No data sent to servers
• Fast - Predictions in real-time
• Offline-capable - Works without internet

Model Architecture¶

The suggestion system uses:

• TensorFlow Lite - Optimized for mobile
• NLP Model - Natural language processing
• Icon Classifier - Visual recognition
• Category Predictor - Context-aware categorization

Training Data¶

Model trained on:

• Common grocery items
• Shopping list patterns
• Item-category relationships
• Icon-name associations

Item Name Suggestions¶

Auto-complete¶

As you type, see suggestions:

1. Start typing (e.g., "milk")
2. See suggestions: "Milk", "Milkshake", "Milk chocolate"
3. Tap to select or keep typing

Context-Aware¶

Suggestions consider:

• Your purchase history
• Current list category
• Time of year (seasonal items)
• Common pairings

Learning From You¶

The model improves with use:

• Frequently used items appear first
• Custom item names remembered
• Store-specific preferences learned

Icon Suggestions¶

Automatic Icon Assignment¶

ShopSync suggests appropriate icons:

• Food items - Relevant food icons from library
• Household - Cleaning, personal care icons
• Custom - Best guess for unusual items

Icon Library¶

Over 500 icons across categories:

• 🥬 Produce (fruits, vegetables)
• 🥛 Dairy products
• 🥩 Meat & seafood
• 🍞 Bakery items
• 🥫 Pantry staples
• 🧼 Household items
• And more...

Custom Icons¶

Don't like the suggestion?

• Tap the icon to browse all options
• Search for specific icons
• Categories organized for easy finding

Category Prediction¶

Smart Categorization¶

Items automatically categorized:

• "Apples" → Produce
• "Cheddar cheese" → Dairy
• "Ground beef" → Meat
• "Paper towels" → Household

Confidence Scores¶

High-confidence predictions auto-assigned. Low confidence suggestions shown for confirmation.

Override Options¶

Don't agree with the category?

• Tap to change
• Choose from category list
• Your choice remembered for future

Quantity Suggestions¶

Common Quantities¶

Suggested based on item type:

• Milk - "1 gallon", "Half gallon", "Quart"
• Eggs - "1 dozen", "18 count"
• Apples - "5 lbs", "1 bag"

Smart Defaults¶

Previous quantities for the same item:

• If you usually buy "2 lbs" of apples
• That becomes the default suggestion

Enabling/Disabling¶

Toggle Smart Suggestions¶

In Settings → Lists & Items:

• Smart Suggestions - Main toggle
• Auto-Categorize - Category predictions
• Icon Suggestions - Automatic icon assignment
• Quantity Presets - Quantity suggestions

Performance¶

Resource Usage¶

Minimal impact on device:

• Model size - ~5 MB
• RAM usage - ~20 MB during predictions
• Battery - Negligible impact
• CPU - Runs on background thread

Prediction Speed¶

• Typing latency - <50ms
• Icon lookup - <100ms
• Category prediction - <100ms

Privacy & Data¶

On-Device Processing¶

All predictions happen locally:

• ✅ No data sent to servers
• ✅ Your items stay private
• ✅ Works offline
• ✅ GDPR compliant

Data Collection¶

ShopSync does NOT:

• ❌ Upload your shopping lists
• ❌ Track individual items
• ❌ Sell your data
• ❌ Use your data for ads

Model Updates¶

Models updated periodically:

• Downloaded with app updates
• Optional manual model download
• Backwards compatible

Future Enhancements¶

Upcoming ML features:

Price Predictions¶

Estimate costs based on typical prices.

Shopping Patterns¶

Learn your shopping habits:

• "You usually buy milk on Sundays"
• "Time to restock toilet paper"

Recipe Integration¶

Add recipe ingredients automatically.

Store Layout Learning¶

Optimize list order for your favorite store.

Troubleshooting¶

Suggestions Not Appearing¶

1. Check if enabled - Settings → Smart Suggestions
2. Update model - Settings → Advanced → Update ML Model
3. Clear cache - May need to relearn patterns
4. Reinstall app - Last resort

Poor Suggestions¶

1. Use items more - Model improves with usage
2. Manually categorize - Teaches the model
3. Provide feedback - Report suggestion issues

High Battery Usage¶

If smart suggestions use too much battery:

1. Check background refresh is disabled
2. Update to latest app version
3. Disable if not needed

Technical Details¶

Model Specifications¶

• Framework - TensorFlow Lite
• Input - Text string (item name)
• Output - Top 5 predictions with confidence scores
• Latency - <100ms on modern devices
• Model size - 4.8 MB

Implementation¶

Key components:

• SmartSuggestionsService - Manages ML model
• FoodIconDetector - Icon prediction
• ItemCategoriesService - Category assignment

See Services Documentation for technical details.

Best Practices¶

Next Steps¶

• Working with Items
• Categories
• Services Guide