Wayne Choy@TheHonouredOne

Our Track 1 submission is a batch AI agent engineered for the competition's scoring rule: clear the accuracy gate, then win on fewest tokens. Since only tokens routed through the Fireworks proxy are counted, we make most tasks free. The agent is local-first: a bundled Qwen2.5-3B-Instruct model (4-bit GGUF) runs on CPU via llama.cpp and answers each of the eight task categories — factual, math, sentiment, summarization, NER, code debugging, logical reasoning, and code generation — at zero Fireworks tokens. A zero-token heuristic router assigns each prompt a category, prompt template, output cap, and verifier. Conservative per-category verifiers keep well-formed local answers and escalate only clear failures to a Fireworks model as a safety net. On our 40-task benchmark the local model alone reaches 90% accuracy. Everything ships as one self-contained, publicly pullable linux/amd64 Docker image.

To participate in the hackathon, simply click Sign up with AMD. If you are not already a member of the AMD AI Developer Program (ADP), you will need to create an account. ADP members can access AI tutorials, experts, and community support. In addition, new members can claim a free month of DeepLearning.ai pro, $100 in AMD GPU cloud credits, and $50 in Fireworks AI API credits. All hackathon participants will receive on hackathon launch additional compute and API credits for their projects. Ready to build what's next on AMD? Sign up with AMD below to secure your spot.