PosturePro

Motivation Isn't Enough

People aren't struggling with motivation alone; they're struggling with confidence in how they move. Through user interviews with beginner gym-goers and independent exercisers, three barriers showed up again and again:

1. Lack of confidence exercising alone — Beginners can't tell if their form is safe or effective, creating hesitation and drop-off.
2. Limited access to guidance — YouTube and fitness apps demonstrate movements but never tell you if your posture is correct. Personal trainers solve that, but at $60–120/session most people can't sustain them.
3. Real injury risk from poor posture — Without in-the-moment feedback, beginners repeatedly load joints incorrectly, leading to strains, back pain, and exercise-related injuries.

Confidence in Movement Is the Real Product

Across interviews, two patterns crystallized:

• Videos and apps are one-way: they show the "ideal" but never react to your actual form.
• Trainers are interactive but financially and logistically inaccessible for the average beginner.

The real unmet need wasn't "more content." It was interactive feedback on my body, right now, delivered in a way that feels accessible and non-judgmental.

An AI "Mini-Trainer" in Your Pocket

PosturePro uses your smartphone camera plus AI to monitor your posture and give coaching cues while you exercise.

• The app analyzes body position during exercises like squats, lunges, and push-ups using on-device pose detection (MediaPipe, MoveNet, OpenPose).
• It detects form issues (e.g., knees collapsing inward, excessive spine flexion, uneven weight distribution).
• It translates those detections into simple coaching cues like "Shift your weight back onto your heels" or "Keep your chest more upright."

The goal is not to replace personal trainers entirely. PosturePro is a lightweight AI mini-trainer focused narrowly on posture and injury prevention at a fraction of the cost.

Proving Value Before Building the Whole Thing

Instead of jumping directly into full real-time video, I scoped an MVP that tested the riskiest assumption cheaply: Is AI posture feedback actually accurate and useful to users?

MVP Workflow

• User uploads a photo of themselves doing an exercise
• AI analyzes posture from the still image
• System returns posture corrections and coaching cues

This allowed fast iteration on the core value (feedback quality and trust) before investing in real-time streaming and latency optimization.

How It Works

I designed the system to deliver fast, privacy-preserving, human-readable coaching.

1. On-device pose detection — MediaPipe Pose, MoveNet, OpenPose output body landmarks (shoulders, elbows, hips, knees, spine alignment). Raw video stays on-device for latency and privacy.
2. Posture evaluation — Compute joint angles and alignments against biomechanical movement standards. Detect common issues like knee cave, excessive spinal curvature, or asymmetric weight distribution.
3. LLM-based coaching — Pose detection yields structured signals ("knee angle too narrow," "forward spine tilt"). An LLM converts those into natural coaching cues.
4. Audio coaching (TTS) — Coaching cues run through text-to-speech so users can hear feedback mid-set, similar to a trainer talking beside them.

Experiments: Does This Work and Will Anyone Pay?

I treated PosturePro like a real product, not just a demo, and explicitly tested two core assumptions.

Experiment 1 — Posture Detection Accuracy

Hypothesis: AI-generated posture feedback from smartphone images will be perceived as accurate and useful by beginners.

Method: Prototype built using Replit + Gemini Vision. 23 beginner gym users uploaded photos of themselves doing squats, push-ups, and lunges. The system generated posture corrections and users rated them.

The prototype validated that AI feedback is both technically feasible and perceived as valuable, clearing the path to invest in real-time posture monitoring.

Experiment 2 — Willingness to Pay

Hypothesis: Users will pay $20/month for an AI posture coach framed as a "mini-trainer."

Method: Simulated landing page and pricing page. Users could choose to subscribe at $20/month or opt out, with follow-up questions on why.

Unit Economics

From the start, I designed PosturePro to behave like a real SaaS business, not a one-off app.

• Pricing: $20–25/month subscription, positioned as "a fraction of the cost of a trainer."
• Revenue per user: $20–25/month
• Variable cost per user: ~$5–6/month (on-device inference amortized, backend, LLM, TTS, payments, lightweight support)
• Contribution margin: ~70%+, leaving room to fund acquisition while staying healthy

From B2C Wedge to Platforms

I mapped out a three-phase expansion path:

Phase 1 — B2C Wedge

Target beginners and rehab patients exercising independently. Channels include physiotherapy clinics referring patients at discharge and fitness communities where form anxiety is high.

Phase 2 — Smart Gym Infrastructure

Enable gyms to become "AI-enabled" without buying new hardware. Deploy PosturePro Smart Stations — tablets or kiosks that offer posture feedback in existing gym spaces.

Phase 3 — Corporate Wellness & Networks

Extend PosturePro where musculoskeletal injuries are a cost center: corporate wellness programs, insurance providers, physiotherapy and rehab networks.

How I Worked

I led end-to-end discovery and definition as Product Manager:

• Designed and ran user research with beginner exercisers to uncover the emotional layer behind "I want to work out more."
• Framed the problem as confidence and injury risk, not just "better workouts."
• Defined the product scope, positioning, and wedge segments (rehab → beginners → gyms → enterprise).
• Collaborated on AI prototype development using pose detection + LLM + TTS architecture.
• Designed and ran experiments on accuracy and willingness to pay, with explicit success thresholds and decision rules.
• Built the business model, unit economics, and staged go-to-market plan across B2C, smart gyms, and corporate wellness.

What Our Advisor Said