Why I'm Building an AI-Matched Fitness Marketplace

Finding a personal trainer online is broken. I know because I've been on both sides of it.

The Problem Nobody Talks About

Every major fitness platform treats trainer discovery like a catalogue. You get a grid of faces, a handful of checkboxes — "speciality: weight loss", "location: London" — and a prayer that the person you pick actually understands how you train.

That's not matching. That's browsing.

Meanwhile, the trainers on those platforms are paying for ads, boosting Instagram posts, and hoping the algorithm surfaces them to the right people. The average solo personal trainer spends more time marketing than coaching. That's backwards.

I've watched talented coaches burn out — not because they couldn't deliver results, but because they couldn't find the clients who needed them. And I've watched motivated clients bounce between three or four trainers before finding someone who actually got them. The discovery layer is the bottleneck, and nobody is fixing it properly.

Why Filters Don't Work

Here's the core issue: fitness goals are nuanced. "I want to lose weight" could mean a hundred different things. It could mean a postpartum mother rebuilding core strength. A desk worker with chronic back pain wanting to move without fear. A former athlete who wants to feel competitive again.

Checkbox filters collapse all of that into "weight loss." Then they match you with whoever ticked that box.

Semantic matching is different. It reads the meaning behind what you describe — your goals, your history, your preferences, your constraints — and finds the trainer whose profile, experience, and coaching style genuinely align. Not the trainer who selected the right dropdown option. The trainer who fits.

That's what PumplAI does. We use vector embeddings (pgvector with HNSW indexing, for the technically inclined) to map trainer profiles and client goals into the same semantic space. The match isn't based on keywords. It's based on meaning.

The AI That Stays on Your Device

Here's the part most people don't expect: the AI that generates your workout plans runs on your device. Not in the cloud. Not on our servers.

We use Qwen3, an open-source language model, running locally on your device via Ollama — not on our servers, not in the cloud. Your training data, your injury history, your goals — none of it leaves your device unless you explicitly choose to share it with your trainer.

Why does this matter? Because fitness data is health data. And health data deserves better than being uploaded to a server farm in Virginia so a company can train their next model on your knee injury.

We're EU-based, EU-hosted (Supabase in Frankfurt), and GDPR-compliant by architecture — not just by policy document.

What Trainers Actually Need

I've talked to dozens of personal trainers building their businesses online. The pattern is consistent:

They don't need another tool to manage clients. Trainerize, TrueCoach, Everfit — the management layer is solved. What's not solved is the acquisition layer.

Solo trainers earning $2,500–$6,000 a month spend a disproportionate amount of energy finding their next client. They post content on Instagram, run paid ads, rely on word-of-mouth, and cross their fingers. The platforms they use are built for retention, not discovery.

PumplAI flips that. Clients come to you, matched by AI based on genuine fit — not because you outbid someone on an ad platform.

And the business model reflects that: subscription-based, recurring revenue for trainers. Not hourly billing that caps your income at the hours in your day. A subscription model where 20 engaged clients at $100/month means $2,000/month of predictable income — without chasing leads.

Form Checking That Actually Works

One more thing we're building that nobody else has: real-time form checking using computer vision.

MediaPipe Pose runs in the browser via TensorFlow.js — no app install, no server processing. Point your camera at yourself during a squat, and the system gives you real-time feedback on depth, knee tracking, and spine position.

This isn't a gimmick. It's the feature that makes PumplAI something clients show their friends. Trainers use it as a coaching tool between sessions. Clients use it as a safety net when training alone.

No competitor has this. ABC Trainerize doesn't. TrueCoach doesn't. Everfit doesn't. FirstRep doesn't. The closest thing in the market is generic "AI trainer" apps that bypass human coaches entirely — which defeats the purpose.

Why Now, and Why Me

I'm a solo founder building this from the EU. I don't have a team of 50. I have five AI agents that help me ship code, write content, manage infrastructure, and run QA. It's a different way to build, and it's working.

The fitness coaching market is at an inflection point. Trainers are going online faster than the tools are keeping up. Clients are demanding personalisation that generic apps can't deliver. And the incumbents are focused on enterprise gym chains, not the solo trainer grinding on Instagram.

The gap is specific: a two-sided marketplace where AI handles the matching, a human handles the coaching, and the technology stays out of the way.

That's PumplAI. AI-matched, human-coached.

Join the Waitlist

We're opening early access to trainers and clients in the coming weeks.

Trainers: If you're tired of chasing clients and want a platform that brings them to you, matched by AI — join the trainer early-access list →

Clients: If you want a personal trainer who actually fits your goals, not just whoever shows up first in a directory — join the client waitlist →

The right trainer for you exists. AI finds them. They coach you.

---

PumplAI is built in the EU, hosted in the EU, and designed for privacy from the ground up. Your data stays yours.