Blog
Biorce has won Best Poster in the Drug Development category at ESMO AI 2026. The ESMO AI & Digital Oncology Congress jury recognized our findings showing how clinical-purpose AI outperforms general-purpose LLMs in drafting oncology protocols.
Drafting clinical trial protocols is a rigorous, time-consuming process that demands strict adherence to regulatory and ethical frameworks. Because general models lack clinical trial architectures, we tested a simple question: could our clinical-purpose AI, Aika, generate more complete protocols from a basic prompt under blinded expert review? Here is what we found:

Our research findings
To ensure an unbiased benchmark, we randomly selected 10 Phase II/III interventional oncology trials from ClinicalTrials.gov and fed the exact same prompt, built from each trial's registry brief summary, into four systems: Aika, Claude Opus 4.8, ChatGPT GPT-5.5, and Gemini 3.6. Three blinded oncology Subject Matter Experts (SMEs) then scored all 40 protocols on a 0-to-5 scale across SPIRIT 2025 items 9–34, yielding the following results:
Aika significantly outperformed general-purpose models, achieving 73% overall SPIRIT 2025 compliance compared to 21.5% to 57% for the general-AI assistant.
Aika generated the longest, most consistent drafts, whereas general-purpose outputs varied widely and left entire sections unaddressed.
Aika scored 4.1–4.3/5 on vital operational items, including confidentiality, retention, adherence, and data management, compared to 0.8–1.7 on average for the other models.
A moment to celebrate at ESMO AI
At ESMO AI, we had inspiring conversations with researchers from academia, pharma, and biotech during our poster presentation. In addition, a second Biorce study was selected and defended in a rapid oral session, marking another major milestone for our team. We are truly thankful to the ESMO AI scientific committee for recognizing our research and highlighting the importance of clinical-specific AI in driving more reliable and efficient clinical trials.








