Bison Chat

Overview

Bison Chat is a teaching demo built for Lipscomb's College of Computing — a proxy that routes prompts between Anthropic's Haiku, Sonnet, and Opus based on the prompt's complexity. The goal is to make model selection visible to students learning about LLM economics.

What students see

• They type a prompt
• The router classifies it (lightweight question, code task, deep reasoning task, etc.)
• They watch the proxy pick a model, justify the pick, and call it
• The response and the model choice are both visible side by side

Why it works as a teaching tool

When students start using LLMs, they default to "the smartest one" and run up bills. Watching the proxy correctly send a fact lookup to Haiku and a multi-step reasoning task to Opus makes the cost/capability tradeoff land in a way no slide ever does.

Stack

• Python proxy server
• Anthropic SDK with all three models
• A simple chat UI showing prompt, router decision, and response together