Sentedel

Secure what's
private.

Purpose-built models for detecting and redacting PHI in healthcare EDI -- where general-purpose tools fall short.

View Benchmarks Read v1 Release Public playground
Relaxed Recall (50% Overlap)
100%
Strict Accuracy (Exact Boundary)
91.0%
Performance

Unmatched PHI Detection
on EDI Data

General-purpose PII models fail at parsing X12 formatting. Sentedel v1 is fine-tuned specifically for healthcare EDI, achieving perfect recall.

Model Strict Recall Relaxed Recall (50%) Strict F1
Sentedel EDI-PHI v1 		91.0% 100.0% 76.4%
GLiNER PII Base 49.0% 54.2% 50.3%
NVIDIA GLiNER PII 35.9% 40.9% 39.8%
OpenAI Privacy Filter 3.0% 64.4% 1.2%

Zero Missed PHI

Across 13 PHI categories and 500 test transactions, v1 caught 11,000 out of 11,002 elements (99.98%).

Boundary Precision

S-tag label mapping fixed subword tokenization errors, pushing exact strict matches from 0.8% to 91%.

Hyper-Optimized

Dynamic sequence packing and LoRA optimization reduced training time by 62%.

Latest Research

Technical Blog Benchmark

PHI Detection for EDI Transactions.

by Sentedel Research June 2026

Building the first PHI detection benchmark for payer-side healthcare EDI — and why general-purpose models fail dramatically on structured transaction data.

91.0% STRICT
100% relaxed recall
Read the post