AMD AI Agent & Video Captioning Suit

This submission combines two AMD Developer Hackathon tracks into a single AI project. Track 1 implements a Hybrid Token-Efficient Routing Agent capable of solving a wide range of natural language tasks including factual question answering, mathematical reasoning, text summarization, sentiment analysis, named entity recognition, logical reasoning, code generation, and debugging. The agent reads tasks from an input JSON file, intelligently routes requests to appropriate Fireworks AI models, and produces structured JSON output while minimizing token usage for improved efficiency. Track 2 implements an automated Video Captioning Agent. The application downloads the input video, extracts representative frames, analyzes the visual content using AI vision models, and generates captions in four different styles: Formal, Sarcastic, Humorous Tech, and Humorous Non-Tech. The solution is designed to generalize across different video categories rather than relying on predefined examples. Both tracks are fully containerized using Docker, developed in Python, and organized in a modular architecture for maintainability and scalability. The project uses modern AI APIs, structured processing pipelines, and reusable components to deliver reliable and efficient performance while satisfying the AMD Developer Hackathon requirements.