Team Idea
Solo builder in the AMD AI League, building practical AI systems that run on AMD Instinct GPUs with ROCm.
Match 3: a containerized RAG pipeline that indexes a folder of mixed documents (PDF, Word, Excel, CSV, logs, source code and images) and answers each question with the exact value plus the exact files it came from. One vision-language model (Qwen3-VL-8B) reads text inside images and writes the answers; hybrid BM25 + dense retrieval follows ticket and part IDs across files for multi-hop questions; strict grounding rules refuse anything the documents don't state.
Next up: a web information-retrieval agent (Match 4), multi-agent code repair (Match 5) and a token-efficient game agent (Match 6). Code will be published open source as each Match closes.
Team Members
Caleb Samante
caleb_samante755
Submission
Submission draft in progress