How Giles AI Achieves 95% Accuracy in Medical Literature Extraction with Gemini

Bringing a new drug to market typically takes 10 to 12 years, with roughly a third of that time consumed by literature review and evidence synthesis. Researchers must sift through databases like PubMed — which hosts over 40 million articles — to identify the specific evidence needed for a clinical trial or regulatory submission. The work is time-consuming, error-prone, and deeply consequential: a missed study or misread result can delay a therapy by years.

Giles AI was founded to compress this timeline. Its core product, giles®, is a research assistant designed to think like a healthcare professional — ingesting scientific literature and extracting structured, cited insights that researchers can trust. Trust is the critical constraint in healthcare AI: hallucinations that might be tolerable in other contexts are unacceptable when the output informs clinical decisions. Giles AI’s previous cloud provider could not supply the compliance infrastructure or the specialized AI capabilities the company needed, requiring it to stitch together multiple third-party solutions. The architectural fragmentation made HIPAA and SOC2 compliance significantly more difficult to achieve and maintain.

Migrating to Google Cloud with support from partner Insight, Giles AI rebuilt on a unified infrastructure. Vertex AI became the foundation for model orchestration, allowing the team to swap between Gemini models for complex reasoning and open-source models like Gemma for enterprise clients with data residency requirements. Document AI parses unstructured scientific literature — clinical studies, regulatory submissions, patents — and feeds structured outputs to Gemini Pro for analysis and synthesis. If the information needed isn’t present in the source material, the system explicitly says so rather than generating a plausible but false answer. This behavior was critical to gaining trust from clinical researchers.

The results are measurable. Giles AI now achieves 95% accuracy in medical data extraction. One clinical customer reduced the time required for research tasks by 85%. The AI’s outputs agree with those of human researchers at a rate of 98%. Google Kubernetes Engine and Cloud Run underpin the infrastructure layer, significantly reducing latency so that features like text-to-speech work instantaneously. The company achieved healthcare compliance out of the box, removing a significant barrier to enterprise adoption.

Giles AI is now expanding the platform from a research assistant toward a real-time collaborator capable of listening in on research meetings via Gemini Live, proactively surfacing relevant literature in context. Plans include integration with MedGemma and TxGemma for medical image analysis, and a launch on Google Cloud Marketplace to reach enterprise clients in Europe and Latin America without procurement friction. Giles AI CEO Rishi Wadhera estimates the 18 months of development the company has completed would have taken significantly longer without Google Cloud’s integrated infrastructure and direct team support.