3CoffeeNoLunch

Our submission is an autonomous AI agent that intelligently routes tasks to the most cost-efficient Fireworks AI models while maintaining strict accuracy standards (≥80% accuracy gate). The agent combines a hierarchical routing system with real-time cost optimization to minimize token consumption across diverse task categories. Core Features: Adaptive Router: Classifies incoming tasks (factual Q&A, math, sentiment analysis, code debugging, NER, logic, summarization) and selects the optimal model from Fireworks' available endpoints using cost-per-token analysis Hybrid Inference Strategy: Leverages quantized local models (2B-3B parameter) for low-complexity tasks alongside Fireworks API calls, achieving zero-token scoring on confident local predictions Structured Logging & Telemetry: Complete tracing of routing decisions, model selections, and cost metrics for analysis and optimization Containerized Deployment: Docker image with linux/amd64 manifest, optimized for the standardized evaluation environment