STRUGGL: Zero-LLM-Math AI coach for hybrid athletes

Comments

0

Taha BazizMaker

What features should STRUGGL add to improve hybrid athletic coaching?

About STRUGGL

STRUGGL is the training system for hybrid athletes balancing heavy strength training, running, and Hyrox/ATHX race preparation — built on a strict Zero-LLM-Math architecture where language models never guess physiological numbers.

Instead of letting an LLM hallucinate working weights, heart-rate zones, or race splits, STRUGGL routes every physiological calculation through deterministic, peer-reviewed sports-science engines: - Strength & autoregulation: Wathan 1RM + RPE/RIR working-load solver with full bodyweight-kinematic accounting for weighted calisthenics (pull-ups, dips, muscle-ups) - Running & cardio: Tanaka HRmax + Karvonen Heart-Rate Reserve (HRR) Zone 2 prescriptions - Hypertrophy ledger: Fractional 17-muscle weekly hard-set tracking against MEV / MAV / MRV landmarks - Concurrent fatigue & guardrails: 7:28 exponentially weighted ACWR, same-day interference spacing, and joint-pain plan gates - Compromised Hyrox pacing: Per-lap running + 8-station split solver calibrated against real race fatigue decay

Available on Web, iOS, and Android — plus an open-core Remote MCP connector (OAuth 2.1 + PKCE) and Claude Agent Skills so athletes and builders can audit, plan, and log training directly inside Claude.

How it was built with Claude

Remote MCP Server + MCP Apps UI (struggl.singandclap.com/mcp): Built a 12-tool Streamable HTTP MCP server with OAuth 2.1 + PKCE dynamic client registration (readOnlyHint / destructiveHint annotations on every tool) and an interactive MCP Apps HTML volume & readiness card rendered inline in chat. Stateless tools (hyrox_pace, strength_benchmark) work immediately with zero login; personal ledger tools (get_profile, get_sessions, get_muscle_volume, plan_workout, log_workout) trigger OAuth 2.1 only when needed.
Zero-LLM-Math Tool Boundary: Prompts and SKILL.md workflows (struggl-coach and fitness-agent-primitives) forbid the model from doing mental arithmetic on loads, paces, or HR zones — forcing structured tool calls into deterministic Python physiology engines.
Claude Code Subagents (fitness-math-auditor & struggl-hybrid-coach): Used specialized subagents to audit exercise-physiology edge cases (calisthenics total system mass, EWMA vs rolling-boxcar ACWR spikes, concurrent lower-body interference gates) across our backend and API tests.