From Prompt to Proof: Governing LLM Intelligence in Pharma with Snowflake and Shiny
R/Medicine 2026
Tanya Cashorali, TCB Analytics
This talk presents a production-ready AI architecture for generating defensible competitive and clinical intelligence in the pharmaceutical industry. While large language models can summarize biomedical content, healthcare decisions demand traceability, source attribution, and reproducibility. We demonstrate how to transform LLM outputs into governed, auditable, and reusable insights aligned with real-world regulatory and strategic requirements.
Our approach separates probabilistic AI reasoning from deterministic data execution. The intelligence pipeline lives natively in Snowflake and is exposed through REST API endpoints, making it fully language agnostic. R, Python, or other clients can interface with the same services while keeping the user interface decoupled from orchestration and core logic.
We showcase LLM-based question classification, structured entity extraction, semantic caching with embeddings, knowledge graph execution for grounded retrieval, and source-aware integration with PubMed and ClinicalTrials.gov. A Shiny interface deployed with Posit Team demonstrates how human-in-the-loop validation, audit trails, and versioned knowledge objects can support institutional memory and continuous monitoring. Attendees will learn reproducible design patterns for building trustworthy, scalable AI systems in regulated healthcare environments.
Resources
R/Medicine: https://rconsortium.github.io/RMedici...
R Consortium: https://www.r-consortium.org/