← Back to Work
Insurance
Insurance Underwriting AI
Led data science for an AI system that improved document extraction accuracy by 40%.
Document AI ML Engineering Data Science
Key Outcomes
- — 40% improvement in extraction accuracy
- — Reduced manual review time by 60%
- — Scaled to process 10K+ documents daily
The Challenge
Insurance underwriting relies on extracting accurate information from complex, unstructured documents—applications, medical records, financial statements. Manual processing was slow and error-prone.
The Approach
At Sixfold, I built and led the ML team that developed document understanding systems. We combined computer vision, NLP, and domain-specific rules to extract structured data from messy documents.
Technical Highlights
- Designed multi-modal document understanding pipeline
- Fine-tuned transformer models for insurance-specific extraction
- Built confidence scoring system for human-in-the-loop review
- Implemented active learning to continuously improve models
The Outcome
The system achieved 40% improvement in extraction accuracy, enabling faster underwriting decisions and better risk assessment.