ClinAgent: A ReAct-Based Agent for Conversational Access to Clinical Trial Information
| Source: arXiv AI
Tags: clinical trials, RAG, ReAct, biomedical AI, Gemini, DeepSeek, agentic AI
ClinAgent, accepted at CIBB 2026, lets clinicians query clinical trial registries in plain language using a ReAct-based LLM agent — with Gemini 3.0 Flash achieving the highest overall performance and DeepSeek V3.2 thinking mode excelling at planning quality.
Details
Searching clinical trial registries like ClinicalTrials.gov remains manual and error-prone, requiring familiarity with structured query syntax. ClinAgent introduces a conversational RAG system using the ReAct paradigm, allowing users to query trial information in natural language across multi-turn sessions. The system integrates three tools: a ClinicalTrials.gov search interface, a PubMed module, and a Python-based analyzer operating on a locally cached structured dataset. Evaluation used a three-phase framework covering operational effectiveness, planning quality, tool-use efficiency, and expert qualitative judgment. Three LLM backends were compared: Gemini 3.0 Flash and two DeepSeek V3.2 variants (thinking and non-thinking). DeepSeek thinking mode excelled in planning quality; Gemini 3.0 Flash achieved the highest overall performance and expert ratings. The complementary strengths suggest hybrid backend strategies for production deployments. Accepted at CIBB 2026, the paper addresses a genuine clinical workflow gap. Local caching of structured trial data supports offline use — relevant for clinical environments with constrained connectivity or data residency requirements.