.png&w=256&q=75)
š¤ Latest Submissions
.png&w=828&q=75)
Hybrid Token Efficient Routing Agent
Hybrid Token Efficient Routing Agent is a Dockerized AI agent built for Track 1 of the AMD Developer Hackathon ACT II. The project focuses on completing a fixed set of tasks while using the fewest Fireworks AI tokens possible. The system first analyzes each prompt, estimates its difficulty, detects the task category, and attempts to solve it locally using deterministic logic. Simple arithmetic, conversions, date helpers, string operations, text extraction, JSON checks, greetings, and basic factual prompts can be handled without calling a remote model. This gives those tasks zero Fireworks token cost. If the local solver cannot answer with high confidence, the router selects the cheapest suitable Fireworks model based on task type, difficulty, prompt length, and ambiguity. Fireworks API communication is isolated in a dedicated client with retries, timeout handling, exponential backoff, response validation, and structured logging. The project is containerized with Docker, configured entirely through environment variables, and designed for deterministic execution, token efficiency, maintainability, and clear evaluation.
13 Jul 2026
