Hyperlex is a SaaS solution for contract management. Since its creation, Hyperlex has been developing its own artificial intelligence. It is thanks to this AI that we can offer our service. It analyses contracts and extracts the important data they contain to facilitate your daily work.
It is a complex technology and an entire team is dedicated to its daily development and improvement.
We are very proud of this work and would like to highlight it in a series of articles dedicated to their expertise. You will find them in the Tech section.
In order to better understand the writings they will be publishing, we offer you this first article dedicated to the organization of Hyperlex's Machine Learning team. May it help you to understand the specificities of their job and their research.
What is Artificial Intelligence?
Artificial intelligence includes all the techniques that allow us to draw inspiration from human thought in order to create machines capable of relieving humans of complex tasks.
There are two types of AI:
- symbolic AI, that is, AI that only acts according to the rules dictated by humans,
- AI by learning, that is to say the one that tries to make sense of the input data, to detect the signals present in these data, and that can learn to recognize them. This is called Machine Learning.
It is this last discipline that is mainly used at Hyperlex.
Machine learning and data
Machine Learning is the set of techniques that allow to identify correlations (or "patterns") in data sets in order to derive predictive models.
Then, from these models, the machine will be able to analyze unknown documents. This is called prediction.
Without data, there are no models! That's why data is essential to the work of our Machine Learning team.
At Hyperlex, the data we process is the content of contracts. This is why the team needs legal experts. They are the ones who indicate the specificities and constraints related to their expertise.
How is machine learning applied at Hyperlex?
In order to recognize what the contracts contain, an OCR (Optical Character Recognition) system is first applied. This allows us to convert an image into editable text.
💡 Did you know? Today, you can also use this kind of technology on some smartphones! It allows you to copy and paste a text contained in an image.
After this step, the team uses a form of Machine Learning called Natural Language Processing (NLP). It consists in processing and analyzing raw text (of any type of document) and drawing conclusions from it. In short, it can understand the content of this text, locate certain elements of interest or identify recurring sequences.
It is thanks to him that our contract management solution recognizes the characteristics contained in your contracts and accompanies you at each stage of their life cycle!
This diagram helps to understand how it works.
🔎 Going further: our ML team doesn't just use NLP, but also goes through image analysis. You can find out more in this article:
How does image analysis work in contract management?
What does the Machine Learning team look like?
Hyperlex's Machine Learning team has a mission: to improve and evolve its artificial intelligence in order to offer an ever more efficient service!
This translates into the development of new features such as the recognition of more and more languages.
To do this, the team follows a 4-step cycle process:
- Data Acquisition. The team starts by working with Legal Experts to annotate the data, but also with the product team to define the users' needs. This data is then analyzed for quality control using an in-house tool. Then, it is stored in what is called a Data Lake.
- Model training. It consists in recovering the data from this lake, in order to train several models. These models allow us to feed our technology and make it more and more efficient.
- Model evaluation, which consists of testing the reliability of the best model selected. This new model is then compared with the existing model to verify its quality. It must be better on historical and recent data.
- Deployment and monitoring, which consists of monitoring all the machine's predictions in order to identify possible regressions and then taking into account the users' feedback.
💡 Did you know. When you specify the type of clause you are reviewing in Hyperlex, you are helping to improve our machine. Our team takes this feedback into account!
This complex research and iteration work is carried out by 4 groups within the Machine Learning team and under the responsibility of Romain Vial.
The role of the Core Model and Business Model in the Machine Learning team 🧠
Alaa, Fuqi, Ahmed and Hicham are Data Scientists. Their job is to:
- design business solutions by combining different models;
- imagine new architectures to improve existing models;
- develop reusable models for different business problems.
The role of the ML Ops in the Machine Learning team 🧑💻
Akim, Subaandh and Amine are Data Engineers. They are responsible for the model creation process, during all stages of machine learning.
It is the ML Ops team that provides the necessary tools to deploy all the functionalities related to Machine Learning. They are then responsible for the stability of these services.
The role of the Data cluster in the Machine Learning team 💻
Estelle is a Data Analyst. She is in charge of analyzing the gathered data and controlling their relevance and quality.
In addition, it identifies "holes in the racket" by spotting clauses or attributes that the models do not handle correctly. This allows it to restart the annotation and improve the models.
The Tech Lex section
As you can see, developing artificial intelligence is a complex technology. We want to familiarize our readers and/or users with it.
Each month, an article will be dedicated to the latest technology developed by Hyperlex and written by our employees. Experts in their field, they are eager to share what they do on a daily basis.
Simpler content, like this one, will also be available.
Whether you are simply curious, want to join the Hyperlex team or understand our internal operations, the articles in our new Tech Lex section are dedicated to you.
You want to follow Hyperlex more closely?
Subscribe to our newsletter!