OncoGraph: Multi-Agent AI for Precision Oncology

Precision oncology requires synthesizing complex genomic data, medical imaging, and clinical guidelines—a task that is overwhelming and time-consuming for human oncologists. OncoGraph solves this critical challenge by deploying a collaborative multi-agent AI architecture designed specifically for cancer treatment planning. Our system utilizes four specialized AI agents working in parallel. Agent 0 leverages Google Gemma 4 31B running on AMD MI300X GPUs to perform multimodal digital pathology analysis. Agents 1, 2, and 3 utilize GLM 5.2 Fast via Fireworks AI to conduct genomic mutation analysis, clinical pharmacology evaluation, and final treatment synthesis. By integrating these models, OncoGraph automatically identifies key mutations (e.g., EGFR, KRAS, PIK3CA), evaluates drug interactions, and generates comprehensive, evidence-based treatment plans aligned with NCCN/ESMO guidelines. This project demonstrates the power of combining cutting-edge multimodal LLMs with high-performance AMD hardware to accelerate personalized medicine and improve patient outcomes.