Kshitij Kumar@kshitijkumarkksunny99577

Token Router is built on one simple rule: never pay for tokens that a cheaper solver can earn for free. Every task flows down a four-step ladder, cheapest first. Pure arithmetic gets parsed and solved exactly by deterministic Python, costing zero tokens. Sentiment, NER, factual Q&A, summarization and math word problems run on a Qwen2.5-3B model on CPU inside the container via llama.cpp, still zero Fireworks tokens. Only the genuinely hard work escalates: logic puzzles and low-confidence local answers go to minimax-m3, while code generation and debugging go to kimi-k2p7-code, both at temperature 0 with tight token caps. Accuracy is protected by design, not hope. The math solver only fires on pure arithmetic prompts, never on word problems that just happen to contain numbers. The reasoning budget for minimax-m3 is sized so answers never come back empty. If Fireworks fails for any reason, the agent falls back to its local answer, so no single task can sink the run. Nothing is cached or hardcoded, every task is solved live. The results back it up: 9 out of 9 sample tasks correct with only 596 Fireworks tokens, 6 of them solved completely free. The full container run scored 6 out of 6 in 30 seconds end to end on CPU alone. The image is linux/amd64, 5.08 GB against the 10 GB cap, publicly pullable, and all secrets come in through runtime env vars with nothing baked into the 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.