Initial Submission

This submission is a token-efficient general-purpose AI agent built in Python and containerized with Docker. It is designed to handle all eight evaluated capability categories: factual knowledge, mathematical reasoning, sentiment classification, text summarization, named entity recognition, code debugging, logical/deductive reasoning, and code generation. The core design principle is minimizing token usage without sacrificing accuracy. Rather than calling an LLM to classify each task, the agent uses a lightweight rule-based keyword classifier that routes each prompt to the correct handler at zero token cost. Each category is paired with a minimal one-sentence system prompt and a calibrated max_tokens cap to prevent over-generation. Model selection is dynamic: the agent parses parameter size from model IDs (e.g. 70b, 8x7b) at runtime and routes simpler tasks — sentiment, NER, factual, summarization — to the smallest available model, while reserving the largest model for computationally harder tasks like code generation, logical puzzles, and multi-step math. All model IDs and API credentials are read from environment variables at runtime as required by the harness. The container reads tasks from /input/tasks.json and writes results to /output/results.json, exiting with code 0 on success.