Clinical data rarely arrives analysis-ready. In a recent project, our team used bfPREPTM to convert 10,000+ pages of clinical PDFs into an OMOP-structured dataset suitable for analysis and machine learning using bfLEAPTM. The organizing idea was simple: use agents as the connective tissue between intake and analysis, and keep a human in the loop wherever judgment, safety or nuance are required.
This session walks through a practical blueprint. By using bfPREPTM, first, the primary parsing task of PDF-to-CSV which covers triage of documents, schema generation and iterative text extraction. Second, tooling for entity resolution to align concepts to OMOP vocabularies (e.g., SNOMED CT, RxNorm, LOINC), with confidence thresholds that trigger human review. Third, templates for agents that edit existing schema-constrained loaders and transformation snippets for OMOP tables. With a verified core in place, we perform feature engineering using OMOP-compatible tools, then extend with agentic assists.
Examples include categorizing diseases, medications, lesion descriptors and normalizing free text into intended categories. Throughout, we prioritize medium-complexity repeatables that a human can verify quickly. We will cover failure modes we encountered, such as clipped layouts and column ambiguity, and the guardrails that kept them contained. Attendees will leave with templates, patterns and decision criteria to stand up agent-powered intake without hype. The goal is not full automation. It is verified automation that moves messy clinical documents into trustworthy features while preserving human judgment, auditability and speed.
Speaker
Juan Felipe Beltrán Lacouture, PhD, Director of AI, Machine Learning & Innovation, BullFrog AI
Juan Felipe Beltrán, PhD, is Director of AI, Machine Learning and Innovation at BullFrog AI, where he builds explainable ML platforms for clinical trials and drug discovery. He specializes in human-in-the-loop systems that transform complex biomedical data into production-grade analytics, with experience processing thousands of clinical samples from brain banks and multi-omics studies. He holds a PhD in Computational Biology from Cornell University and has published in Science, Nature Methods and other leading journals.
Who Should Attend?
This webinar will be of interest to:
- Clinical Data Managers and Study Data Leads
- Biostatisticians and Statistical Programmers
- Data Scientists and ML Engineers in biopharma and CRO settings
- Clinical Informaticians
- OMOP and OHDSI Practitioners
- RWE, HEOR and Pharmacovigilance teams
- Data Engineering, ML Ops and Platform Leaders
- Product Managers and Technical Leaders building data platforms
What You Will Learn
Attendees will gain insights into:
- A step-by-step PDF-to-OMOP pattern that uses agents for speed and humans for judgment
- Safe use of agents for intake script editing and feature engineering within constraints
- Early detection and mitigation of common intake failure modes
Xtalks Partner
BullFrog AI
BullFrog AI leverages Artificial Intelligence and machine learning to advance drug discovery and development. Through collaborations with leading research institutions, BullFrog AI uses causal AI in combination with its proprietary bfLEAP™ platform to analyze complex biological data, aiming to streamline therapeutics development and reduce failure rates in clinical trials.
You Must Login To Register for this Free Webinar
Already have an account? LOGIN HERE. If you don’t have an account you need to create a free account.
Create Account