Naman Varshney
Back to Blogs

Building HYROX Coach: The First Station-by-Station AI Training Platform

Sep 20, 2025•6 min read•
AILLMKafkaRAGHYROXWearablesFitness

Train Smarter. Track Every Station.

HYROX Coach is the first AI-powered training platform that connects with Apple Watch & Garmin to record every station split, pace, and heart rate — then builds personalized, data-driven plans that evolve as you train.

I built this out of frustration. Existing fitness apps treat HYROX like a long run. There was no way to track stations, analyze pacing, or see where I broke down in a simulation. So I built the tool I needed to train smarter.

From 103→77 kg to first HYROX finish — I'm turning hard-won lessons into an AI coach that understands your stations, not just your splits.

The Problem: Why Generic Fitness Apps Fail HYROX Athletes

Picture this: You finish a HYROX simulation. Your watch says you ran 8 km and your heart rate averaged 165 bpm. But here's what you don't know:

  • Which sled station broke you? Was it the push or the pull?

  • Did your wall ball pace drift? By how much? When?

  • What was your recovery between stations? Did you come into the row too hot?

  • Where did you actually lose time? The watch shows splits, but not station performance.

This was my reality after every training session. I'd finish a simulation, look at my watch data, and have no idea what actually happened. The data was there—heart rate zones, pace, splits—but it was useless for HYROX.

Existing apps treat HYROX like a long run. They track distance and pace, maybe heart rate zones, but they have no concept of:

  • Station-specific performance

  • Transition times between run and work

  • Station breakdown patterns

  • Recovery metrics between stations

I needed something that understood HYROX as 8 runs + 8 stations, not just "8 km with some exercises."

The Solution: What HYROX Coach Actually Does

Here's how it works in practice:

🏃‍♂️ Station-by-Station Tracking

Before: Your watch records a 1 km run, then you manually remember "I did sled push here."

With HYROX Coach: You tap your watch at the start of each station. The app records:

  • Sled Push: Time, distance, heart rate drift, perceived effort

  • Run segments: Pace, HR recovery, transition time

  • Wall Balls: Reps, rest breaks, HR spikes, form breakdown points

  • Every station: Complete performance profile

Real example: After a simulation, you see: "Sled Push #2 took 45 seconds longer than #1, and your HR didn't recover below 160 bpm before the next run. This is where you broke down."

📈 Adaptive AI Plans

The problem with generic plans: They assume you respond the same way to the same stimulus. But HYROX training is station-specific.

How HYROX Coach adapts:

1. Week 1: You complete a simulation. The AI analyzes your station performance.
2. Week 2: Based on your data, it adjusts:
- "Your sled push form breaks down after 3 sets. Let's add tempo sled work."
- "Your wall ball pace drifts after 50 reps. We'll build sets of 25 with controlled rest."
- "Your run pace is solid, but HR spikes on transitions. Let's work on breathing resets."

3. Week 3: The plan evolves again based on new data.

It's not just "do more sleds." It's "do more sleds this way because your data shows you break down here."

🧠 Simulation Builder

The scenario: You want to test a race strategy. Maybe you're planning to go harder on the first 4 stations, then hold steady.

With HYROX Coach:
1. Build a custom workout: "4K Power Gauntlet" — 4 km run with 4 stations (sled push, sled pull, burpees, row)
2. Set target paces and station times
3. Execute on your watch with station-by-station tracking
4. Get instant feedback: "You went 15 seconds too fast on sled push #1, which caused your row split to drop by 8 seconds. The strategy needs adjustment."

Test before race day. Know exactly how your body responds to different pacing strategies.

📊 Performance Analytics

What you get after every session:

  • Station breakdown: Time, HR drift, recovery, pacing for each station

  • Pattern recognition: "Your sled performance drops after station #4. This is consistent across 3 simulations."

  • Recovery metrics: "You need 90 seconds between sled push and run to keep HR below 165. Currently averaging 60 seconds."

  • Power output estimates: Based on sled load, distance, and time

  • Comparative analysis: "This week's wall ball pace was 12% faster than last week, with 5 bpm lower HR. Progress."

No more guessing. See exactly where you need to improve.

💪 Built by a HYROX Competitor

I finished HYROX Mumbai in 1:37:23. I know what it's like to:

  • Finish a simulation and have no idea what happened

  • Try to analyze watch data that doesn't understand HYROX

  • Follow generic plans that ignore station-specific weaknesses

  • Train without knowing if you're actually improving

HYROX Coach is the tool I wish I had during my prep. It's built for athletes who want precision, not guesswork.

How It Works: The Technical Architecture

The Pipeline: Watch → Kafka → Feature Store → LLM Planner → UI

Data Flow

1. Watch Integration: Apple Watch & Garmin sync station markers, heart rate, pace, and splits in real-time
2. Event Streaming: Data flows through Kafka for reliable, scalable processing
3. Feature Store: Station performance, recovery metrics, and patterns are stored and indexed
4. AI Planner: An LLM analyzes your data and generates personalized training recommendations
5. UI Dashboard: You see station breakdowns, trends, and adaptive plans

Why This Architecture?

Reliability: End-to-end 99.9% uptime target with automatic retries. Your training data never gets lost.

Intelligence: The LLM planner uses:

  • Guardrails: Ensures recommendations are safe and effective (no "do 200 wall balls tomorrow")

  • Unit tests: Yes, we test the AI's outputs to ensure quality

  • Function-calling: Structured station blocks that the AI can reason about

  • Context windows: Budgeted to keep costs reasonable while maintaining quality

Performance: Cached embeddings and optimized queries mean you get insights in seconds, not minutes.

Scalability: Built to handle thousands of athletes training simultaneously.

What's Next: The Roadmap

Phase 1: Native Watch App (Q1 2026)


  • Automatic station detection: The watch recognizes when you start a station (no manual taps needed)

  • Real-time coaching: Haptic feedback and audio cues during training

  • Offline mode: Track complete simulations without phone connectivity

Phase 2: Advanced Analytics (Q2 2026)


  • Threshold estimation: More accurate power and pace zones from field tests

  • Predictive modeling: "Based on your current fitness, you'll finish in 1:32:00 ± 2 minutes"

  • Injury risk detection: Early warning signs from movement patterns and HR drift

Phase 3: Community & Competition (Q3 2026)


  • Leaderboards: Compare station performance with other athletes

  • Training groups: Share workouts and compete in challenges

  • Coach mode: Coaches can monitor multiple athletes' progress

Why This Matters

HYROX is unique. It's not running. It's not CrossFit. It's 8 runs + 8 stations, and each station requires different skills, pacing, and recovery.

Generic fitness apps can't handle this complexity. They treat HYROX like a long run with some exercises thrown in. But that's not how you train for HYROX, and it's not how you race HYROX.

HYROX Coach is built for athletes who want to:

  • Know exactly where they break down

  • Train with precision, not guesswork

  • See measurable progress week over week

  • Race with confidence, knowing their strategy is tested

Join the Waitlist

HYROX Coach is currently in private beta. If you're training for HYROX and want station-by-station tracking and adaptive AI planning, join the waitlist to get early access.

For athletes who want precision, not promises.

---

Tagline: "Track. Analyze. Adapt. HYROX Coach is the only platform that understands your stations — not just your splits."

Built by a HYROX competitor who finished Mumbai in 1:37:23 and wanted better tools for the next race.