🤓 Latest Submissions
SXR Routing Agent
This is a general-purpose AI agent built for Track 1 of the AMD Developer Hackathon: ACT II. It reads a set of tasks from /input/tasks.json and produces answers in /output/results.json, handling eight distinct capability categories: factual knowledge, mathematical reasoning, sentiment classification, text summarisation, named entity recognition, code debugging, logical reasoning, and code generation. The core design goal is token efficiency. All inference runs through the Fireworks AI API using an allowed model, with the agent preferring Gemma models to also compete for the Best Use of Gemma challenge. Each incoming task is classified into its category using lightweight, zero-token pattern matching, and then given a tailored, minimal instruction and a category-specific output cap. This keeps both prompt and completion tokens as low as possible while protecting answer quality, since the accuracy gate must be cleared before token ranking applies. The agent also robustly handles reasoning-style models by extracting the final answer from their output, and includes automatic fallback across allowed models so a single model failure cannot break a run. The whole solution is containerized with Docker and reads all credentials and model IDs from environment variables injected at evaluation time.
13 Jul 2026
