PhysioAgent is Physiotarget’s multi-agent AI decision-support system designed for oncology
Guideline-Grounded. PhysioAgent uses structured SBOC (Brazilian Society of Clinical Oncology) guidelines and related documents as its primary clinical knowledge sources. It delivers recommendations as structured clinical decisions while maintaining retrieval context.
State-of-the-art Deep Reasoning. PhysioAgent uses the open-weight Google DeepMind Gemma 4 26B MoE model, fine-tuned to produce responses aligned with SBOC guidelines.
Offline and On-Premises. PhysioAgent runs efficiently on affordable NVIDIA hardware. The RAG (Retrieval-Augmented Generation) pipeline stores local copies of guideline sources and uses BioBERTpt, a Portuguese biomedical and clinical language model, for local retrieval. This eliminates the need for external communication during normal operation. Patient data remains within the hospital, and the system complies with Brazil’s LGPD.
Safety Guardrails. A secondary model, Qwen3Guard, reviews prompts and responses for personally identifiable information, violent or illegal content, and other unsafe categories.
Flexible Solution. PhysioAgent’s document base can include additional documents at the client’s request, and our proprietary LLM fine-tuning methodology enables fast fine-tuning of the Gemma 4 model to meet the client’s needs.