LLM-Router-Agent
This image contains a lightweight AI agent built for the AMD ACT II Developer Hackathon Track 1. The agent reads tasks from /input/tasks.json, processes each prompt using Fireworks AI through the provided FIREWORKS_BASE_URL, and writes valid answers to /output/results.json. It is designed to handle multiple task types, including factual questions, math reasoning, sentiment classification, summarization, named entity extraction, code debugging, logical reasoning, and code generation. The implementation uses a simple deterministic controller with task analysis, category-based planning, Fireworks API execution, and basic output cleanup. It reads FIREWORKS_API_KEY, FIREWORKS_BASE_URL, and ALLOWED_MODELS from environment variables at runtime, so it can run inside the official evaluation harness without hardcoded credentials or model IDs. The goal is to prioritize correctness first while keeping prompts compact and token usage efficient.
Category tags:
.png&w=96&q=75){kind=small}
For Innovators & Creators