1a493e0ad9
feat: recipe engine — assembly templates, prep notes, FTS fixes, texture backfill
...
- Assembly template system (13 templates: burrito, fried rice, omelette, stir fry,
pasta, sandwich, grain bowl, soup/stew, casserole, pancakes, porridge, pie, pudding)
with role-based matching, whole-word single-keyword guard, deterministic titles
via MD5 pantry hash
- Prep-state stripping: strips 'melted butter' → 'butter' for coverage checks;
reconstructs actionable states as 'Before you start:' cooking instructions
(NutritionPanel prep_notes field + RecipesView.vue display block)
- FTS5 fixes: always double-quote all terms; strip apostrophes to prevent
syntax errors on brands like "Stouffer's"; 'plant-based' → bare 'based' crash
- Bidirectional synonym expansion: alt-meat, alt-chicken, alt-beef, alt-pork
mapped to canonical texture class; pantry expansion covers 'hamburger' from
'burger patties' etc.
- Texture profile backfill script (378K ingredient_profiles rows) with macro-derived
classification in priority order (fatty → creamy → starchy → firm → fibrous →
tender → liquid → neutral); oats/legumes starchy-first fix
- LLM prompt: ban flavoured/sweetened ingredients (vanilla yoghurt) from savoury
- Migrations 014 (nutrition macros) + 015 (recipe FTS index)
- Nutrition estimation pipeline script
- gitignore MagicMock sqlite test artifacts
2026-04-02 22:12:35 -07:00
3016efa65b
fix: address recipe/OCR quality issues from review
2026-04-02 12:41:59 -07:00
22e57118df
feat: add DocuvisionClient + cf-docuvision fast-path for OCR
...
Introduces a thin HTTP client for the cf-docuvision service and wires it
as a fast path in VisionLanguageOCR.extract_receipt_data(). When CF_ORCH_URL
is set, the pipeline attempts docuvision allocation via CFOrchClient before
loading the heavy local VLM; falls back gracefully if unavailable.
2026-04-02 12:33:05 -07:00
33a5cdec37
feat: cloud auth bypass, VRAM leasing, barcode EXIF fix, pipeline improvements
...
- cloud_session.py: CLOUD_AUTH_BYPASS_IPS with CIDR support; X-Real-IP for
Docker bridge NAT-aware client IP resolution; local-dev DB path under
CLOUD_DATA_ROOT for bypass sessions
- compose.cloud.yml: thread CLOUD_AUTH_BYPASS_IPS from shell env; document
Docker bridge CIDR requirement in .env.example
- nginx.cloud.conf + nginx.conf: client_max_body_size 20m for barcode uploads
- barcode_scanner.py: EXIF orientation correction (PIL ImageOps.exif_transpose)
before cv2 decode; rotation coverage extended to [90, 180, 270, 45, 135]
to catch sideways barcodes the 270° case was missing
- llm_recipe.py: CF-core VRAM lease acquire/release wrapping LLMRouter calls
- tasks/runner.py + config.py: COORDINATOR_URL + recipe_llm VRAM budget (4GB)
- recipes.py: per-request Store creation inside asyncio.to_thread worker to
avoid SQLite check_same_thread violations
- download_datasets.py: HF_PARQUET_FILES strategy for repos without dataset
builders (lishuyang/recipepairs direct parquet download)
- derive_substitutions.py: use recipepairs_recipes.parquet for ingredient
lookup; numpy array detection; JSON category parsing
- test_build_flavorgraph_index.py: rewritten for CSV-based index format
- pyproject.toml: add Pillow>=10.0 for EXIF rotation support
2026-04-01 16:06:23 -07:00
9371df1c95
feat: recipe engine Phase 3 — StyleAdapter, LLM levels 3-4, user settings
...
Task 13: StyleAdapter with 5 cuisine templates (Italian, Latin, East Asian,
Eastern European, Mediterranean). Each template includes weighted method_bias
(sums to 1.0), element-filtered aromatics/depth/structure helpers, and
seasoning/finishing-fat vectors. StyleTemplate is a fully immutable frozen
dataclass with tuple fields.
Task 14: LLMRecipeGenerator for Levels 3 and 4. Level 3 builds a structured
element-scaffold prompt; Level 4 generates a minimal wildcard prompt (<1500
chars). Allergy hard-exclusion wired through RecipeRequest.allergies into
both prompt builders and the generate() call path. Parsed LLM response
(title, ingredients, directions, notes) fully propagated to RecipeSuggestion.
Task 15: User settings key-value store. Migration 012 adds user_settings
table. Store.get_setting / set_setting with upsert. GET/PUT /settings/{key}
endpoints with Pydantic SettingBody, key allowlist, get_session dependency.
RecipeEngine reads cooking_equipment from settings when hard_day_mode=True.
55 tests passing.
2026-03-31 14:15:18 -07:00
0d65744cb6
feat: StyleAdapter — 5 cuisine templates with element dimension biasing
2026-03-31 12:54:42 -07:00
3943a8c99d
fix: grocery_links — guard against empty ingredient names in build_links
2026-03-31 12:44:58 -07:00
e8fb57f6a2
feat: RecipeEngine Level 1-2 — grocery links + affiliate deeplink builder
...
Add GroceryLink schema model and grocery_links field to RecipeResult.
Introduce GroceryLinkBuilder service (Amazon Fresh, Walmart, Instacart)
using env-var affiliate tags; no links emitted when tags are absent.
Wire link builder into RecipeEngine.suggest() for levels 1-2.
Add test_grocery_links_free_tier to verify structure contract.
35 tests passing.
2026-03-31 12:23:07 -07:00
37737b06de
feat: RecipeEngine Level 1-2 — corpus match, substitution, grocery list, hard day mode
2026-03-31 11:50:28 -07:00
96850c6d2a
feat: SubstitutionEngine — deterministic swap candidates with compensation hints
2026-03-30 23:13:49 -07:00
e57ae74e27
fix: staple library — consistent tofu_firm slug, load error handling, typed yield_formats, expanded test coverage
2026-03-30 23:10:51 -07:00
a03807951b
fix: ElementClassifier — guard empty input, safe JSON decode, dedup heuristic elements, strengthen test assertions
2026-03-30 23:10:49 -07:00
e377bd85aa
feat: ElementClassifier -- ingredient element tagging with heuristic fallback
2026-03-30 22:59:46 -07:00
