AI Wellness App
Logging meals, workouts and weight is the friction that kills most wellness apps. This MVP reduces it to a chat-like input with photo and barcode capture, confirmation cards, dashboards and coaching workflows. RemSoft.Dev specified and built the applied-AI core: OpenAI-backed structured parsing, multimodal food and label analysis, deterministic-first parsing with GPT fallback, an AI coach using tool calling with persistent profile facts and summary memory. AI uncertainty was turned into product requirements - output validation, suspicious-input checks, item caps, macro-calorie consistency flags, timeouts, fallback behaviour, per-user usage limits and mocked-OpenAI testing rules. A documented 500-case QA improvement cycle raised the pass rate from 15% to 83.8%; a two-tier vision approach - low-detail food photos, high-detail label images - cut estimated vision costs roughly sevenfold.