← Back to Features

ML Name Detection

Machine learning detection of person names, locations, and organizations using a pre-trained NER model that runs entirely in your browser.

Privacy First

All machine learning processing happens 100% in your browser. Your data never leaves your device. The ML model is downloaded once and cached locally in your browser's IndexedDB storage.

No server. No cloud. No data collection.

What It Detects

👤 Person Names

Full names, titles, and name variations that pattern-based rules might miss. E.g., "Dr. Sarah Johnson", "Steve Bull"

📍 Locations

Cities, countries, regions, and place names. E.g., "London", "Silicon Valley", "New York"

🏢 Organizations

Company names, institutions, and organizations. E.g., "Microsoft", "NHS", "University of Cambridge"

Technology

ML Name Detection uses state-of-the-art Named Entity Recognition (NER) technology:

Transformers.js - Hugging Face's JavaScript library for running transformer models in the browser
BERT Models - Bidirectional Encoder Representations from Transformers, fine-tuned for NER
ONNX Runtime - Efficient model execution via WebAssembly
IndexedDB Caching - Model persists across browser sessions

Available Models

Model	Download Size	Speed	Accuracy	Best For
DistilBERT NER	~250 MB	Fast	Good	General use (recommended)
BERT Base NER	~420 MB	Slower	Best	Maximum accuracy
BERT Base NER (uncased)	~420 MB	Slower	Best	Case-insensitive text

How to Use

Enable ML Detection

Open the sidebar and expand the Extras section. Check the ML Name Detection checkbox.

Select a Model

Choose from DistilBERT (faster, smaller) or BERT Base (more accurate, larger). DistilBERT is recommended for most use cases.

Download the Model

Click Download Model. This only happens once — the model is cached in your browser for future sessions.

Analyze Your Text

Click Analyze. ML detection runs automatically alongside pattern-based rules, finding names that regex can't catch.

Review & Scrub

Check the Smart Suggestions panel for ML-detected entities. Click Scrub to anonymize everything.

When to Use ML Detection

ML detection is most valuable when:

Processing free-form text with person names not in standard formats
Need to detect organization or company names
Location names need to be identified and scrubbed
Pattern-based rules are missing names in narrative text
Working with customer support tickets or comments

Limitations

False positives - May occasionally flag non-names as entities
False negatives - Unusual names or contexts may be missed
Language - Models are trained primarily on English text
Performance - Larger models are slower; use DistilBERT for speed

Ready to try ML Name Detection?

No installation required. Open LogScrub, enable ML detection, and start analyzing.

Launch LogScrub

All Features • Pattern Matching • Documentation