Adaptive AI Model Router

Most AI applications send every request to the same model, regardless of how complex it actually is — wasting cost and latency on questions that don't need a powerful model at all. Adaptive AI Model Router solves this with an intelligent routing layer that classifies each incoming prompt (simple / medium / complex) using lightweight heuristics, then routes it to the model tier that actually fits: a fast, cheap model for simple queries, and a powerful model reserved for genuinely complex reasoning. Users aren't locked into the router's judgment — a live priority toggle (Speed / Cost / Quality / Balanced) lets them override the routing decision on the fly, and the same question demonstrably gets sent to a different model depending on what's selected. Every routed request is logged with its classification, latency, token usage, and cost, building a real dataset for future model-based routing decisions. The system is a small, working stack end to end: a Streamlit frontend, a FastAPI backend exposing a /route endpoint, the routing/classification logic, three Groq-hosted LLMs behind it, and SQLite logging. It's built to progressively grow into a full AI/ML systems project — the next phase replaces the hand-written routing rules with a model trained on the logged usage history.