Inspiration As a BMX rider and fitness enthusiast, I saw the need for an AI coach that could provide expert guidance on BMX techniques, fitness training, and equipment selection. The challenge was creating an intelligent system that could autonomously retrieve and reason with specialized knowledge.
What it does AI Coach Bot provides expert coaching on BMX, fitness, and product knowledge through an agentic AI system. It autonomously searches through 65+ curated articles, retrieves relevant information, and uses advanced reasoning to provide contextual responses with source citations. Users can ask about BMX techniques, training routines, or equipment recommendations.
How we built it Built using NVIDIA's llama-3.1-nemotron-nano-8b-instruct for reasoning and nvidia/nv-embedqa-e5-v5 for semantic retrieval. The RAG pipeline processes user queries through embedding-based search, retrieves relevant context, and generates informed responses. Deployed on AWS EKS with Docker containers, featuring a Flask backend and interactive web frontend.
Challenges we ran into Integrating NVIDIA NIMs with the RAG pipeline required careful API management and fallback mechanisms. Optimizing semantic search performance while maintaining response quality was challenging. AWS EKS deployment complexity with proper resource allocation and cost management within the $100 credit limit.
Accomplishments that we're proud of Successfully implemented true agentic behavior with autonomous retrieval and reasoning. Achieved seamless integration of NVIDIA NIMs with AWS infrastructure. Created a scalable, production-ready system that provides accurate, source-attributed responses. Built comprehensive deployment automation for easy replication.
What we learned Gained deep experience with NVIDIA NIM microservices and their integration patterns. Learned advanced RAG optimization techniques and AWS EKS deployment strategies. Discovered the importance of fallback mechanisms and robust error handling in production AI systems.
What's next for AI Coach Bot Expand knowledge base to include video content analysis and real-time coaching feedback. Implement multi-modal capabilities for image-based equipment recommendations. Add personalized training plan generation and progress tracking features.
Log in or sign up for Devpost to join the conversation.