🏋️‍♀️ Team Silvera × HoloCoach

Your AI voice trainer that lets you keep your real coach - wherever you go.

---

🧠 Overview

Team Silvera × HoloCoach was built during HuddlehiveHack25 as an AI-powered voice trainer designed to extend, not replace, real fitness coaches.
Our mission is to make workouts hands-free, adaptive, and personal — while keeping the connection between trainers and clients alive.

Whether you’re at home, in the gym, or travelling, you can train with your own coach — or choose from a library of diverse coach personas like motivational Gen Z, calm mindfulness, or military precision.

🗣️ “Your AI voice trainer that lets you keep your real coach — wherever you go.”

---

🚀 Features

🎙️ Voice-Driven Training

• Start and control your workout entirely by voice:
  • “Start exercise”
  • “Lower intensity”
  • “Skip set”
  • “Break”
  • “Emergency”

💪 Adaptive AI Coaching

• The AI adjusts workout intensity based on user feedback and session progress.
• Coaches can license their voice and routines — allowing their style and motivation to reach users anywhere.
• Supports real coach cloning using local voice synthesis tools (planned).

🧘 Safety & Personalisation

• AI ensures safe and realistic exercises based on validated routines.
• Adaptive rest timing, intensity adjustments, and progressive tracking keep sessions effective and injury-free.

📊 Social Motivation

• Leaderboards, daily streaks, and progress analytics turn consistency into a game.
• Users stay connected through shared community challenges and friendly competition.

---

🏗️ Architecture Overview

⚛️ Frontend – React + Vite + Web Speech API

• Voice command recognition (always listening)
• Displays exercise state, progress, and voice shortcuts
• Simple and responsive UI for live workout feedback

🧩 Backend – Node.js + Express + Zod

• Handles workout flow as a finite-state machine
• Validates all AI outputs for consistency and safety
• Endpoints:
  • POST /session → Start or reset workout
  • POST /agent/step → Process user command
  • GET /status → Return current workout state

🧠 AI Layer – Ollama + Llama 3.2

• Local AI model for decision-making (no external APIs)
• Uses policy prompts to interpret user intent (e.g., “log reps”, “start rest”)
• Responds in structured JSON only, validated by backend schemas
• Example output:

  {
    "decision": "tool",
    "assistant_say": "Logged 10 reps of Push-ups. Set 1 complete.",
    "tool": "log_rep",
    "args": { "reps": 10 }
  }