Crafting an AI-Based Recommendation System for Lawyers

How Manz built a recommendation system based on semantic document similarity


© [Hubertl]( / Wikimedia Commons / CC BY-SA 4.0


Lawyers, judges, and other legal professionals build a well-rounded picture of a case by looking at related incidents, making sure that no precedent escapes their attention. But depending on the complexity of the case, this can mean leafing through hundreds or even thousands of documents – a significant time investment.

While existing natural language processing (NLP) models speed semantic search, the language of law is structured so differently from ordinary speech and writing — leveraging legal expressions, foreign words, abbreviations, and modified sentence structure — that there is no readily-available universal model that caters to the legal profession.

But Manz, a legal publishing house based in Austria, set out to change that. Leveraging deepset Cloud, Manz built a NLP-driven recommendation system based on semantic document similarity – to speed research workflows, and reduce effort and the associated costs.


Manz’s online legal database, RDB Rechtsdatenbank, is a paid offering that assists legal professionals in Austria. Home to over three million documents, from legal rulings and contractual clauses, to other published content, it is updated on a daily basis. Using RDB database and the German BERT language model as a foundation, Manz leveraged deepset Cloud to craft a powerful semantic search-based system that returns thirty high-quality documents similar to the one the user has searched.

To provide the data needed for fine-tuning the German BERT model to the legal domain, three professionals from Manz labeled 11,000 data points. Once the data was annotated, Manz started building their NLP solution in deepset Cloud. The team continuously refined the legal language model in a two-step process: first using a larger, publicly available dataset by LAVIS-NLP, then training on a subset of the Manz data points.

Integrating the model into RDB was straightforward. “If you have a little bit of experience with software development, then it’s going to work,” says Alexander Feldinger, product manager at Manz.


Ever since this new feature was added to RDB in early November, users have been able to benefit from a much better experience thanks to the new document similarity search. Upon viewing a document, users now receive recommendations for thirty semantically related documents from Manz’ vast collection of legal documents. What’s more, the similarity-based recommendation system returns far more pertinent results than an earlier competitor’s implementation — though both have access to the same pool of data.

Improve your products with NLP

Keen to learn more? Read our Manz case study and discover just how NLP has impacted their business.

Are you looking for a solution that lets you handle the NLP development cycle with confidence and care, allowing you to keep track of the bigger picture, while never losing sight of all the important details? Have a look at deepset Cloud, our enterprise SaaS solution for building NLP backend services quickly and efficiently, making monitoring, retraining, and updating NLP application architectures a breeze.