OnSafety™ Crash Converter

The OnSafety™ Crash Converter is a proprietary AI-powered OCR service that automates data extraction and validation from paper and non-electronically transmitted crash forms. 

Crash Converter consumes both typed and handwritten reports and converts the crash data into structured datasets that adhere to MMUCC and NHTSA guidelines. Here’s how it works:

We retrieve copies of your original crash reports via a secure REST API, file exchange, or an enterprise file management system.

• Files can be scanned paper PDFs, AcroForm PDFs, typed or handwritten crash forms, and various flattened images (PDF, JPG, PNG, GIF, TIFF).
• Each submission is assigned a unique identifier and metadata record.

Our specially trained containerized AI automatically classifies and extracts field data from the crash form.

• Extracted crash data is validated against configurable rule sets based on MMUCC and state schema definitions
• If confidence or rule consistency falls below a specified threshold, records are flagged for human review through a secure web interface
• Feedback from human reviewers continually improves the OCR model’s performance via reinforcement learning on curated datasets

Validated data are written to relational and analytical databases for downstream use. Extracted data can be:

• Sent directly to the State or County’s data repository or warehouse
• Inserted into a normalized “smart” PDF and redistributed
• Exported in a standard format for consumption by any Crash Management System
• XML, JSON, CSV, and Parquet output formats are available and integrate via REST API, secure file transfer, or message queues to partner systems

Secure by Design: All Crash Converter processing occurs within a secure cloud-hosted system.

• The Large Language Models (LLMs) used by our OCR AI are proprietary and explicitly trained on crash form data; they are not available to the public or accessible to other AI models.
• The Crash Converter service uses certificate-based service accounts and secret key management in world-class vault systems.
• PII classifications and tagging are integrated throughout ingestion, validation, and output to enable policy-based masking, retention control, and full lineage tracking, ensuring end-to-end compliance with state privacy and audit requirements.