biome is a data science tool for unstructured data. It provides an effective workflow for training, improving and deploying machine learning classifiers for short, and noisy text such as customer data entries or product data as well as for long texts such as claims or customer service emails. It integrates: 

• Monitoring and analysis tools to help data scientists and business experts to keep track of classification results and improve classifiers over time.
• An easy to use annotation user interface for business experts to create a training dataset from scratch or improve an existing classifier over time.
• Pre-configured and extensible classifiers to let data scientists get started with current best practices while giving them full flexibility to create their own models and data pipelines with Python and PyTorch 

Biome additionally provides biome a set of tools for defining, configuring and training advanced data extraction pipelines with:

• More than 20 out of the box entities such as dates, times, currency amounts, weight, dimensions, and many metric units (like bytes, hertz, decibels, metres, miles, grams, or pounds)
• Support for multiple formats such as PDF, Word, Excel Spreadsheets, HTML, email, or plain text.
• Custom entities, attributes and relation using user defined rules and functions and machine learning models.
• Relational output based on knowledge graphs to let you extract not only entities but their relations, roles and attributes.

The data extraction pipeline supports tabular data such as product catalogues or order spreadsheets and long documents such as customer service emails or product specification PDFs.

Finally, biome provides a service for creating and consuming semantic similarity services with:

• Machine learning for learning a language model out of your own documents.
• Support for multiple formats such as PDF, Word, Excel Spreadsheets, HTML, email, or plain text.
• Analytical user interfaces for finding most similar and dissimilar items in your database or document set at different levels such as record, paragraph or full document-level.