🏆 AI Coach Bot - NVIDIA/AWS Hackathon Submission

An Agentic AI Application built for the NVIDIA/AWS Hackathon featuring advanced RAG (Retrieval-Augmented Generation) capabilities powered by NVIDIA NIM microservices and deployed on AWS EKS.

🎯 Hackathon Requirements Met

✅ llama-3.1-nemotron-nano-8b-instruct large language reasoning model
✅ NVIDIA NIM inference microservice deployment
✅ Retrieval Embedding NIM (nvidia/nv-embedqa-e5-v5)
✅ AWS EKS deployment ready
✅ Complete RAG pipeline with 65+ curated documents
✅ Agentic behavior with autonomous retrieval and reasoning

---

📖 Table of Contents

• About The Project
• Key Features
• Tech Stack
• Quick Start
• Deployment
• Demo
• Architecture
• License
• Contact

---

🚀 About The Project

The AI Coach Bot is an intelligent agentic application that provides expert coaching on BMX, fitness, and product knowledge. Built specifically for the NVIDIA/AWS Hackathon, it demonstrates the power of combining NVIDIA's cutting-edge NIM microservices with AWS's scalable infrastructure.

Key Innovation: The bot autonomously retrieves relevant information from a curated knowledge base of 65+ expert articles and uses advanced reasoning to provide contextual, source-attributed responses.

---

✨ Key Features

• Advanced RAG Pipeline: Semantic retrieval using NVIDIA embedding NIMs
• Expert Knowledge Base: 65+ curated articles on BMX, fitness, and products
• Agentic Behavior: Autonomous information retrieval and contextual reasoning
• Real-time Chat Interface: Interactive web UI with source attribution
• Scalable Architecture: Containerized for AWS EKS deployment
• Performance Monitoring: Token usage and response time tracking
• Fallback Mechanisms: Robust error handling with TF-IDF backup

---

🛠️ Tech Stack

NVIDIA NIMs Used

• LLM: meta/llama-3.1-nemotron-nano-8b-instruct
• Embeddings: nvidia/nv-embedqa-e5-v5

AWS Services

• Amazon EKS - Kubernetes orchestration
• Amazon ECR - Container registry
• Application Load Balancer - Traffic distribution

---

🏗️ Architecture

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   Frontend      │    │   Flask API      │    │  NVIDIA NIMs    │
│   (HTML/JS)     │◄──►│   (RAG Engine)   │◄──►│  LLM + Embed    │
└─────────────────┘    └──────────────────┘    └─────────────────┘
                                │
                                ▼
                       ┌─────────────────┐
                       │  Knowledge Base │
                       │  (65+ Articles) │
                       └─────────────────┘

📂 Project Structure

ai-coach-bot/
├── data/                           # Knowledge base (65+ articles)
│   └── processed_blogs.json
├── scripts/                        # Data processing utilities
├── hackathon_aws_version/          # Main hackathon submission
│   ├── backend/                    # Flask API with RAG pipeline
│   │   ├── app.py                  # Main application
│   │   ├── requirements.txt        # Python dependencies
│   │   ├── Dockerfile             # Container configuration
│   │   └── .env.example           # Environment template
│   ├── frontend/                   # Web interface
│   │   ├── index.html             # Chat interface
│   │   └── script.js              # Frontend logic
│   ├── deploy.yaml                # Kubernetes deployment
│   ├── deploy.sh                  # Automated deployment script
│   └── cleanup.sh                 # Resource cleanup script
└── README.md                      # This file

---

⚡ Quick Start

Prerequisites

• Python 3.9+
• Docker & Docker Compose
• AWS CLI configured
• kubectl and eksctl
• NVIDIA API Key (Get one here)

Local Development

1. Clone the repository

   git clone https://github.com/YOUR_USERNAME/ai-coach-bot.git
   cd ai-coach-bot/hackathon_aws_version

2. Set up environment

   cd backend
   cp .env.example .env
   # Edit .env with your NVIDIA API key

3. Install dependencies

   pip install -r requirements.txt

4. Run locally

   python app.py
   # Open frontend/index.html in browser

Docker Deployment

# Build and run
docker build -t ai-coach-bot:latest backend/
docker run -p 5000:5000 \
  -e NVIDIA_API_KEY=your-key-here \
  ai-coach-bot:latest

---

☁️ AWS EKS Deployment

For complete deployment instructions, see QUICK_DEPLOY.md

One-Command Deploy

cd hackathon_aws_version
./deploy.sh

Manual Steps

1. Create EKS cluster with eksctl
2. Build and push Docker image to ECR
3. Deploy to Kubernetes with kubectl
4. Access via LoadBalancer URL

Estimated Cost: ~$5/day with $100 AWS credits

🎥 Demo

Example Queries

• "What are good BMX bikes for beginners?"
• "How do I improve my cardio for BMX racing?"
• "What should I look for in bike grips?"
• "Explain the basics of a bunny hop"

API Endpoints

• GET /health - System status
• POST /chat - Chat with the bot
• GET /models - Model information

---

🧪 Testing

Health Check

curl http://localhost:5000/health

Chat API

curl -X POST http://localhost:5000/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "What are good BMX bikes for beginners?"}'

📊 Performance Metrics

The system tracks:

• Response time and token usage
• Retrieval accuracy with similarity scores
• Source attribution and document relevance
• Model performance monitoring

🏅 Hackathon Compliance

• ✅ Uses llama-3.1-nemotron-nano-8b-instruct
• ✅ NVIDIA NIM inference microservice
• ✅ Retrieval Embedding NIM (nvidia/nv-embedqa-e5-v5)
• ✅ AWS EKS deployment
• ✅ Complete RAG pipeline
• ✅ Agentic autonomous behavior

📜 License

Distributed under the MIT License. See LICENSE for more information.

📫 Contact

Billy Kennedy - @billykennedybmx - billykennedy@nepa-ai.com

Built with ❤️ for the NVIDIA/AWS Hackathon