Smart-Router Agent

Smart-Router Agent is a token-efficient AI agent built for AMD Developer Hackathon Act II, Track 1. It answers all 8 task categories — factual Q&A, math, sentiment, summarisation, named-entity recognition, code debugging, logic, and code generation — while spending as few paid API tokens as possible. The core is intelligent routing. Every task is sorted by difficulty: the easy ones (facts, sentiment, NER, summaries) are answered instantly by a bundled 2–3B model running locally on CPU, at zero token cost. Only the genuinely hard tasks — math, logic, and code — are sent to the paid Fireworks API, where the extra capability is worth the spend. The result is full-category coverage at a fraction of the token bill. Reliability is built in. If a Fireworks call ever fails, the agent instantly falls back to its local answer, so no task is ever dropped. When every task can be handled on-device, the agent runs entirely free. It's engineered to run anywhere. Smart-Router ships as a lightweight, one-shot Docker container that needs no GPU and fits a modest 4 GB / 2 vCPU box: it reads the task file, routes and answers each item, writes the results, and exits. Deterministic category-based routing keeps memory low, and terse prompting keeps responses fast and cheap. Accurate answers, minimal cost, and portable by design.