A conversational AI fitness app from concept to launch

GloFlow came to Leanware with a concept, no code, and no AI integration. The founder vision was a conversational AI fitness app: a chat-style interface where users talk through their goals, get personalized workout suggestions, photograph their food for recognition and tracking, and watch progress against the goals they set during onboarding. The bet was that the gap between an expensive personal trainer and a generic fitness app could be closed by an LLM-backed assistant that remembered each user context across sessions.

The brief was end-to-end. Mobile app for iOS and Android, conversational AI engine with safety rails for health-related guidance, backend with user management and progress tracking, food-photo recognition pipeline, and the launch readiness work to get the app into both app stores. No part of the system existed prior to the engagement.

The engagement was scoped as AI Product Engineering: a discovery sprint to lock requirements and architecture, followed by a milestone-billed build to ship the product. Fixed budget per phase, with a four-person team of one senior full-stack engineer, two mid full-stack engineers, and a product designer focused on conversational UI.

The mobile app is React Native targeting both iOS and Android from a single codebase. The conversation engine routes through OpenRouter so the platform can switch between models (a cheaper model for casual conversational turns, a more capable model for complex fitness reasoning or food analysis) based on the type of message. Custom prompt and context engineering preserves conversation continuity across sessions: GloFlow remembers what a user worked on last week, what they said about their goals during onboarding, and applies that context to the recommendation it surfaces today.

Health-related guidance flows through an explicit safety layer. The system prompt defines boundaries: GloFlow does not diagnose, does not prescribe, and does not give medical advice. Response filtering catches edge cases that slip past the prompt, and the conversation flow surfaces disclaimers in the appropriate moments rather than burying them in a terms-of-service page nobody reads.

The food-photo feature is its own pipeline. A user photographs a meal, the system runs computer vision against it, identifies the components, looks up nutritional context, and surfaces the result into the conversation. Combined with the smart search (which interprets fitness intent rather than literal keyword matching), the app converts a chat interface into a real fitness tracker without forcing the user to manually log calories.

Backend is Supabase: PostgreSQL for fitness data and user state, Edge Functions for business logic and triggered jobs, Storage for the photos users upload, Auth for sign-in and session management. The stack was chosen for fast time-to-launch with room to handle the operational complexity that comes after users actually arrive.

GloFlow shipped. The app launched on both Apple App Store and Google Play, with the full feature set from the brief in production: conversational AI with cross-session memory, photo-based food recognition, intent-aware search, progress tracking, and the onboarding flow that gives the model the user starting context. The four-month concept-to-launch window held.

The post-launch shape is what the AI Product Engineering line is built for. The founder now has a working product to learn against, the codebase is documented and extensible, and the architecture choices made during the discovery sprint (Supabase for backend, OpenRouter for AI provider abstraction, React Native for cross-platform) leave room for the new features that come out of real user behavior without requiring a re-platform.