Why Incoreperated

Why Incorporated is a general-purpose AI agent built for the AMD Developer Hackathon, designed to handle eight distinct capability domains: factual knowledge, mathematical reasoning, sentiment classification, text summarisation, named entity recognition, code debugging, logical/deductive reasoning, and code generation. Rather than routing every task to a remote model, Why Incorporated runs a lightweight local classifier first, using deterministic pattern matching with an LLM fallback for ambiguous prompts. Each task is then handled by a small local model (Qwen2.5 3B, sized to fit a 4GB RAM / 2 vCPU environment) whenever the category and a multi-signal confidence check (answer validity, structural format matching, and self-consistency across resampled generations) indicate the local model is likely to succeed. Only when local confidence falls below a category-specific threshold does the agent escalate to a Fireworks AI model, selected dynamically from the allowed model list based on live metadata (parameter count, tool support, MoE architecture, coding specialization). For mathematical and code-related tasks, deterministic tool-based solvers run before any model call, resolving simple cases entirely for free. This hybrid design is built around the competition's actual scoring mechanics: local inference is unlimited and free, while every Fireworks token counts against the submission's efficiency rank. Why Incorporated is designed to clear the accuracy gate using local-first inference wherever safely possible, and to fall back to remote reasoning only where it materially improves correctness — treating every remote call as a deliberate, justified spend rather than a default.