NebuFit
Shipped Apr 2026
An AI-powered fitness coaching platform built around a user's real body data — computer-vision physique analysis, RAG-processed bloodwork, and multi-LLM voice coaching — delivered through a production Flutter app with full Stripe subscription management.
Built With
Flutter 
Dart 
Claude API 
GPT-4o 
ElevenLabs 
RAG 
Stripe 
Node.js 
Nest.js 
TypeScript 
PostgreSQL (Supabase) 
Redis Project Overview
NebuFit set out to replace one-size-fits-all fitness templates with coaching that responds to a user's actual body data, bloodwork, and progress — rather than a fixed plan that ignores how an individual actually responds to training. Computer-vision body analysis reads a user's physique directly from photos, while bloodwork gets processed through a RAG system so nutrition and training recommendations stay grounded in real lab results instead of generic advice. The product is built to feel conversational throughout, giving users a coach to talk to rather than another dashboard of static charts.
Underneath that experience, multiple LLM providers — Claude, GPT-4o, and ElevenLabs — are orchestrated together behind an SSE streaming layer, so responses arrive token-by-token in real time instead of after a long wait. Automatic failover between providers keeps voice coaching sessions resilient even if one provider has an outage, which matters for a product where a dropped mid-session call would break trust fast.
The experience is delivered through a Flutter mobile app backed by a Nest.js and TypeScript API, keeping the client and server on a consistent, type-safe foundation. Subscription access and entitlements run on Stripe, so the coaching experience and its monetization shipped together as one production system from day one — giving NebuFit a complete path from AI pipeline to paying subscriber.
- Challenge
- Fitness apps overwhelmingly lean on generic templates instead of adapting to a user's real body composition, bloodwork results, and week-to-week progress — a one-size-fits-all plan doesn't account for someone's actual physiology or how their body responds to training. Users increasingly want conversational, coach-like guidance rather than static charts and spreadsheets: a real answer to a question about a stalled plateau, not just another dashboard. Any solution also had to hold up under the realities of AI infrastructure at scale, since a single LLM provider is a single point of failure, and a coaching product that drops mid-session erodes trust fast.
- Solution
- Built computer-vision body analysis that reads a user's physique from photos, feeding AI-generated workout and nutrition plans that adapt as that data changes. Bloodwork is processed through a RAG system so recommendations are grounded in a user's actual lab results rather than generic advice, with the whole experience delivered conversationally. Multiple LLM providers (Claude, GPT-4o, and ElevenLabs) are orchestrated together behind an SSE streaming layer for real-time, token-by-token responses, with automatic failover between providers keeping voice coaching sessions resilient if one provider drops. Full Stripe subscription and entitlement management is wired in natively rather than bolted on.
- Impact
- Shipped a production-ready AI coaching platform that delivers personalized, real-time guidance grounded in each user's actual body data and bloodwork, rather than generic templates. Resilient multi-provider failover keeps voice coaching sessions from dropping, and Stripe-backed subscriptions mean monetization and entitlement management were built into the platform from day one — giving the product a complete, shippable path from AI pipeline to paying subscriber.
Frequently Asked Questions
Fitness apps lean on generic templates instead of adapting to a user's real body data, bloodwork, and progress, and users want conversational coaching rather than static charts.
NebuFit was built with Flutter, Dart, Claude API, GPT-4o, ElevenLabs, RAG, Stripe, Node.js, Nest.js, TypeScript, PostgreSQL (Supabase), Redis.
This was built as commercial/client work, so the source code isn't publicly available. Get in touch to discuss the implementation.
Computer-vision body analysis reads a user's physique directly from photos, and bloodwork is processed through a RAG system, so workout and nutrition recommendations stay grounded in a person's actual physiology and lab results rather than a one-size-fits-all plan.
NebuFit orchestrates multiple LLM providers — Claude, GPT-4o, and ElevenLabs — with automatic failover between them, so a voice coaching session stays resilient instead of dropping if one provider has an outage.
Responses are streamed token-by-token over an SSE layer, so users see answers arrive live rather than waiting for a full response to generate.
Yes — Stripe subscription and entitlement management is built natively into the platform, so monetization and the coaching product shipped together from day one rather than being bolted on afterward.
Available for work
Got a project like this
in mind?
Let's talk about what you're building and how I can help ship it.