- Rewrite receipt OCR prompt: completes truncated names, preserves fat%
and flavour attributes, extracts weight/volume from line, infers
typical package sizes for solid goods with quantity_confidence field
- Add quantity_confidence to RecognizedItem, EnrichedItem, and
ProductJobResultItem; propagate through item enricher and worker
- Replace per-item create loop with single POST /user-products/batch call
from RecognitionConfirmScreen
- Rebuild RecognitionConfirmScreen: amber qty border for low
quantity_confidence, tappable product name → catalog picker,
sort items by confidence, full L10n (no hardcoded strings)
- Add timestamps (HH:mm / d MMM HH:mm) to recent scan chips
- Show close-app hint on ProductJobWatchScreen (queued + processing)
- Refresh recentProductJobsProvider on watch screen init so new job
appears without a manual pull-to-refresh
- App-level WidgetsBindingObserver refreshes product and dish job lists
on resume, fixing stale lists after background/foreground transitions
- Add 9 new L10n keys across all 12 locales
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add ShelfBarcodeScanScreen: scans barcode via mobile_scanner, looks up
product via GET /products/barcode/{barcode} (Open Food Facts fallback),
returns CatalogProduct to caller; loading overlay while looking up;
"Add manually" fallback in AppBar for unknown products
- Extract AddToShelfSheet to add_to_shelf_sheet.dart (was private in
product_search_screen.dart) so both search and scan screens can reuse it
- Add barcode icon button to ProductSearchScreen AppBar → opens scanner
- Add "Scan barcode" card (📷) to ScanScreen alongside receipt and photo modes
- Rename ScanScreen title: addFromReceiptOrPhoto → scanScreenTitle
("Сканировать" / "Scan & Recognize") to reflect all three modes
- Add 2 L10n keys (scanScreenTitle, barcodeScanSubtitle) across all 12 locales
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Backend:
- Migration 002: product_recognition_jobs table with JSONB images column
and job_type CHECK ('receipt' | 'products')
- New Kafka topics: ai.products.paid / ai.products.free
- ProductJob model, ProductJobRepository (mirrors dish job pattern)
- itemEnricher extracted from Handler — shared by HTTP handler and worker
- ProductSSEBroker: PG LISTEN on product_job_update channel
- ProductWorkerPool: 5 workers, branches on job_type to call
RecognizeReceipt or RecognizeProducts per image in parallel
- Handler: RecognizeReceipt and RecognizeProducts now return 202 Accepted
instead of blocking; 4 new endpoints: GET /ai/product-jobs,
/product-jobs/history, /product-jobs/{id}, /product-jobs/{id}/stream
- cmd/worker: extended to run ProductWorkerPool alongside dish WorkerPool
- cmd/server: wires productJobRepository + productSSEBroker; both SSE
brokers started in App.Start()
Flutter client:
- ProductJobCreated, ProductJobResult, ProductJobSummary, ProductJobEvent
models + submitReceiptRecognition/submitProductsRecognition/stream methods
- Shared _openSseStream helper eliminates duplicate SSE parsing loop
- ScanScreen: replace blocking AI calls with async submit + navigate to
ProductJobWatchScreen
- ProductJobWatchScreen: watches SSE stream, navigates to /scan/confirm
when done, shows error on failure
- ProductsScreen: prepends _RecentScansSection (hidden when empty); compact
horizontal list of recent scans with "See all" → history
- ProductJobHistoryScreen: full list of all product recognition jobs
- New routes: /scan/product-job-watch, /products/job-history
- L10n: 7 new keys in all 12 ARB files
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Rename catalog: ingredient/* → product/* (canonical_name, barcode, nutrition per 100g)
- Rename pantry: product/* → userproduct/* (user-owned items with expiry)
- Squash migrations into single 001_initial_schema.sql (clean-db baseline)
- product_categories: add English canonical name column; fix COALESCE in queries
- Remove product_translations: product names are stored in their original language
- Add default_unit_name to product API responses via unit_translations JOIN
- Add cmd/importoff: bulk import from OpenFoodFacts JSONL dump (COPY + ON CONFLICT)
- Diary: support product_id entries alongside dish_id (CHECK num_nonnulls = 1)
- Home: getLoggedCalories joins both recipes and catalog products
- Flutter: rename models/providers/services to match backend rename
- Flutter: add barcode scan flow for diary (mobile_scanner, product_portion_sheet)
- Flutter: localise 6 new keys across 12 languages (barcode scan, portion weight)
- Routes: GET /products/search, GET /products/barcode/{barcode}, /user-products
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two bugs caused a FK constraint violation on products.unit:
1. RecognizedItem.fromJson fell back to 'шт' (Cyrillic, not a valid
units.code) when the AI returned a null unit — changed to 'pcs'.
2. The unit dropdown in RecognitionConfirmScreen displayed units.keys.first
for invalid units but never updated item.unit, so the invalid value was
still submitted. Added a reconcile step in build() that syncs item.unit
to units.keys.first whenever the stored value is not in the valid set.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace all hardcoded Russian strings in ScanScreen and _LoadingDialog
with AppLocalizations keys; add addFromReceiptOrPhoto, chooseMethod,
photoReceipt, photoReceiptSubtitle, photoProducts, photoProductsSubtitle,
recognizing keys to all 12 ARB files and regenerate AppLocalizations.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add flutter_localizations + intl, 12 ARB files (en/ru/es/de/fr/it/pt/zh/ja/ko/ar/hi),
replace all hardcoded Russian UI strings with AppLocalizations, detect system locale
on first launch, localise bottom nav bar labels, document rule in CLAUDE.md.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Backend:
- Rename recognition_jobs → dish_recognition_jobs; add target_date and
target_meal_type columns to capture scan context at submission time
- Add job_id FK on meal_diary so entries are linked to their origin job
- New GET /ai/jobs endpoint returns today's unlinked jobs for the current user
- diary.Entry and CreateRequest gain job_id field; repository reads/writes it
- CORS middleware: allow Accept-Language and Cache-Control headers
- Logging middleware: implement http.Flusher on responseWriter (needed for SSE)
- Consolidate migrations into a single 001_initial_schema.sql
Flutter:
- POST /ai/recognize-dish now sends target_date and target_meal_type
- DishResultSheet accepts jobId; _addToDiary includes it in the diary payload,
saves last-used meal type to SharedPreferences, invalidates todayJobsProvider
- TodayJobsNotifier + todayJobsProvider: loads unlinked jobs via GET /ai/jobs
- Home screen shows _TodayJobsWidget (up to 3 tiles) between macros and meals;
tapping a done tile reopens DishResultSheet with the stored result
- Quick Actions row: third button "История" → /scan/history
- New RecognitionHistoryScreen: full-screen list of today's unlinked jobs
- LocalPreferences wrapper over SharedPreferences (last_used_meal_type)
- app_theme: apply Google Fonts Roboto as default font family
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove denormalized columns (name, calories, protein_g, fat_g, carbs_g)
from meal_diary. Name is now resolved via JOIN with dishes/dish_translations;
macros are computed as recipe.*_per_serving * portions at query time.
- Add dish.Repository.FindOrCreateRecipe: finds or creates a minimal recipe
stub seeded with AI-estimated macros
- recognition/handler: resolve recipe_id synchronously per candidate;
simplify enrichDishInBackground to translations-only
- diary/handler: accept dish_id OR name; always resolve recipe_id via
FindOrCreateRecipe before INSERT
- diary/entity: DishID is now non-nullable string; CreateRequest drops macros
- diary/repository: ListByDate and Create use JOIN to return computed macros
- ai/types: add RecipeID field to DishCandidate
- Update tests and wire_gen accordingly
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove "Определить блюдо" from ScanScreen and the /scan/dish route.
The + button on each meal card now triggers dish recognition inline —
picks image, shows loading dialog, then presents DishResultSheet as a
modal bottom sheet. After adding to diary the sheet closes and the user
stays on home.
Also fix Navigator.pop crash: showDialog uses the root navigator by
default, so capture Navigator.of(context, rootNavigator: true) before
the async gap and use it to close the loading dialog.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Backend:
- Translate all recognition prompts (receipt, products, dish) from Russian to English
- Add lang parameter to Recognizer interface and pass locale.FromContext in handlers
- DishResult type uses candidates array for multi-candidate responses
Client:
- Add meal tracking: diary provider, date selector, meal type model
- DishResult parser: backward-compatible with legacy flat format and new candidates format
- DishResultScreen: sticky bottom button, full-width portion/meal-type inputs,
КБЖУ disclaimer moved under nutrition card, add date field to diary POST body
- Recognition prompts now return dish/product names in user's preferred language
- Onboarding, profile, home screen visual updates
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add units + unit_translations tables with FK constraints on products and ingredient_mappings
- Normalize products.unit from Russian strings (г, кг) to English codes (g, kg)
- Load units at startup (in-memory registry) and serve via GET /units (language-aware)
- Replace hardcoded _units lists and _mapUnit() functions in Flutter with unitsProvider FutureProvider
- Re-fetches automatically when language changes
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Replace decommissioned llama-3.2-11b-vision-preview with
meta-llama/llama-4-scout-17b-16e-instruct (Groq deprecation)
- Use XFile.readAsBytes() instead of File(path).readAsBytes() so
Android content URIs (from gallery picks) are read correctly
- Add maxWidth/maxHeight constraints to image picker calls to reduce
payload size
- Increase receiveTimeout from 30s to 120s to accommodate slow vision AI
- Log recognition errors via debugPrint instead of swallowing them
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
