Frontend Github link: https://github.com/carolinefve/coach_assistant

Backend/AI Github link: https://github.com/negrahodzic/silvera-holocoach

Presentation Canva link: https://www.canva.com/design/DAG1kPVz2-Y/MGSgdIxOSGC3feFdgkmRgw/edit

FrontEnd Demo link: Team Silvera x HoloCoach x HuddleHive25 x Starling - AI Voice Trainer Frontend

🏋️‍♀️ Team Silvera × HoloCoach

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

📑 Table of Contents

🎯 Framing the Problem
💡 Our Idea
👥 How It Works
🧬 Personalization That Feels Human
🧠 Behind the Scenes — How We Built It
⚙️ Core Architecture Flow
💬 Use Cases
🔧 Technical Highlights
🏆 Accomplishments
🔮 Next Steps
💼 Business Plan & Market Positioning

🎯 Framing the Problem

Fitness isn’t just physical — it’s emotional and relational.
People stay consistent when they feel guided, seen, and supported. Yet today:

• In-person coaching is expensive, limited by location, and not always available when travelling.

• Fitness apps and online videos feel generic — they can’t sense how you’re feeling, adjust intensity, or give real feedback.

• Coaches themselves lose opportunities when clients train remotely.

The result?
Most people struggle to maintain consistent, effective workouts without real-time feedback or personal connection.

💡 Our Idea

Silvera × HoloCoach bridges that gap.

It’s an AI-powered voice trainer that keeps users connected to real human coaches — turning every workout into a personalized, hands-free conversation.

Instead of replacing trainers, it extends their presence:
clients can continue training with their coach’s voice, tone, and motivational style even when apart.

“When your coach can’t be there, Silvera × HoloCoach makes sure their voice still can.”

👥 How It Works

Each session feels like training with your real coach:

You say: “Start exercise.”

The AI greets you and explains today’s plan — for example:
“We’re focusing on biceps today: three exercises, 10 reps each.”

During the workout, you control everything by voice — in your own words.

You don’t need to memorise commands — the AI understands intent naturally.

It uses a local LLM to interpret your speech logically, no matter how you phrase it:

• “Lower intensity” / “Make it easier” → Adjusts the plan

• “Skip this one” / “Next exercise” → Moves forward

• “I need a break” / “Rest a bit” → Starts rest timer

• “Stop for a sec” / “Emergency” → Pauses session safely

After completion, the AI summarises your progress and suggests what’s next — adapting to your energy and preferences.
Every interaction feels natural, motivational, and human — not robotic.

🧬 Personalization That Feels Human

Silvera allows users to choose or import coach personas:

• 🎧 Gen Z motivator — energetic and upbeat

• 🧘 Calm wellness mentor — gentle, balanced tone

• 💪 Military-style trainer — discipline and drive

• 👤 Your real coach — voice-cloned version of your personal trainer

These personas preserve real expertise and emotional authenticity, while AI adapts intensity, pace, and form to each user’s comfort level.

🧠 Behind the Scenes — How We Built It

The system is built around a three-layer architecture designed for flexibility, safety, and responsiveness.

⚛️ Frontend (React + Web Speech API)

• Displays the current workout, progress tracker, and voice-command interface.

• Enables fully hands-free control through natural conversation.

🧩 Backend (Node.js + Express + Zod)

• Manages the finite-state machine for each workout phase: Ready → Active → Rest → Adjust → Complete.

• Validates every AI instruction for safety and coherence.

• Handles real-time updates and syncs with the frontend seamlessly.

🧠 AI Layer (Ollama + Llama 3.2)

• Acts as the decision-maker, parsing user intent (e.g., “add 2 more reps”) and converting it into structured actions.

• Every output is validated using Zod schemas — ensuring no unsafe commands or invalid logic.

Together, these layers create real-time, conversational coaching powered by local AI — no external APIs required.

⚙️ Core Architecture Flow

User → Speech → AI Decision (JSON Output)
      ↓
Backend (Tool Execution)
      ↓
Frontend (Updated State Display)

Example:

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

💬 Use Cases

🔧 Technical Highlights

🏆 Accomplishments

✅ Fully functional voice-driven workout loop
✅ AI decision-to-function pipeline with validation
✅ Scalable policy system for tool actions
✅ Hands-free frontend with visual status + command buttons
✅ Foundation for AI vision and adaptive training

🔮 Next Steps

• 👁️ AI Vision — Posture and form correction

• 🗣️ Coach Voice Cloning — Bring real trainers’ voices into users’ workouts (ElevenLabs)

• 💬 Social Features — Leaderboards, streaks, and community challenges

• 🧩 Adaptive Weekly Planner (Data-Informed AI):
  Built using the dataset provided by HoloCoach, which contains detailed parameters such as:

  • Exercise type and intensity levels

  • Recommended progressions

  • Medical conditions and safe modifications

  • Fatigue and recovery patterns

  In the future, this dataset will be used to train the model so that the AI can reason intelligently about user feedback.
  When a user says something like:

  • “My shoulder hurts,”

  • “I’m feeling fatigue,” or

  • “I have a knee issue,”
    the system will interpret these inputs and adjust the workout plan dynamically — referencing real trainer insightsembedded in the data.

  This means every recommendation is grounded in human expertise, ensuring that AI-guided adjustments remain safe, logical, and medically informed.

💼 Business Plan & Market Positioning

Silvera × HoloCoach transforms how gyms, coaches, and members stay connected beyond physical sessions.
Instead of competing with trainers or selling individual subscriptions, our model partners with gyms and training centers through a simple licensing model.

🏋️ Gym-Integrated Experience

Gyms license the platform and offer it as an exclusive member benefit.
Trainers upload their voice, routines, and motivational style, while members can:

• Train remotely with their coach’s familiar voice, maintaining connection even when travelling.

• Use the app in-gym through earphones to follow form cues and rep counts hands-free.

This creates a seamless extension of real coaching, not a replacement.

💡 Why It Works

• Gyms retain clients and modernize their digital offerings.

• Trainers extend their reach and income without extra workload.

• Users enjoy continuity, motivation, and safe, personal workouts anywhere.

📊 Market Position

Silvera × HoloCoach sits in a unique niche between:

• Traditional fitness apps (too generic)

• Virtual trainers (too detached)

• AI companions (too impersonal)

We combine human coaching authenticity with AI adaptability — a hybrid model the industry is ready for.

Our goal: Make every gym a connected ecosystem where AI amplifies the trainer’s presence, not replaces it.