The Problem
Manual food logging introduces severe input friction. Logging a single meal demands 5 to 10 manual database queries with high variance in portion estimation. Legacy tracking apps operate as passive tabular stores rather than active real-time inference systems.
The Solution
Calibite deploys a multimodal computer-vision pipeline backed by deterministic barcode fallbacks. An image buffer triggers real-time bounding-box inference, volumetric ingredient segmentation, and macronutrient decomposition within ~1.2 seconds.
Key Features
- Real-time bounding-box inference: multimodal vision pipeline detecting ingredient coordinates and volumetric portions within ~1.2s
- Streaming nutritional telemetry: instant macro breakdown streaming via server-sent events with sub-second time-to-first-token
- Deterministic UPC fallback: automated barcode scanning trigger whenever visual inference confidence scores drop below 85%
- 1M+ verified food database: indexed local and cloud nutrition taxonomy with single-document Firestore daily logs
- Adaptive metabolic target engine: dynamic caloric and macronutrient partition modeling based on historical expenditure telemetry
- Fridge-to-Fork pantry pipeline: instant recipe synthesis matching on-hand ingredients against strict macronutrient constraints
System Architecture & Data Pipeline
Client EngineReact Native (Expo managed) unified codebase for iOS & Android, utilizing native device bridges for camera, biometrics, and push.
Vision PipelineCloud Functions proxying GPT-4o Vision requests with image buffer compression (1024x1024) and exponential backoff retry.
Data PersistenceFirestore denormalized daily log schema, collapsing daily meal reads to a single document fetch.
Auth BoundaryFirebase Auth with in-place anonymous-to-authenticated conversion, eliminating upfront registration friction.
In-App PurchasesAdapty SDK for cross-platform Apple/Google IAP, entitlement synchronization, and paywall A/B testing.
CI/CD DeliveryEAS Build and Submit pipelines powering automated App Store and Google Play releases.
Computer Vision Pipeline Flow
[User Camera] ──(Base64/Buffer)──> [Cloud Function Gateway]
│
├──(Compression: 1024x1024 JPEG, Quality 0.85)
▼
[OpenAI GPT-4o Vision API]
│
├──(JSON Schema: FoodAnalysisResult)
▼
[Zod Contract Validation & Macro Normalizer]
│
├──(Confidence >= 0.85) ──> [Firestore Daily Log Document]
└──(Confidence < 0.85) ──> [Trigger UPC Barcode Fallback]
Database Topologies & Collections
users/{userId}
daily_logs/{userId}_{date}
food_items/{itemId}
user_goals/{userId}
iap_entitlements/{userId}
Production Telemetry & Metrics
~1.2s
Vision Inference SLA
1M+
Food Database Entries
142ms
P99 API Latency
iOS + Android
Production Shipped
