The Sifter Factory™
A First-of-Its-Kind Factory
We develop our Sifters® with machine learning and natural language processing to identify a specific concept like a hidden indemnity provision. Our lawyers find hundreds or thousands of examples of a specific concept. Our Data Science team finds the intangible common thread that binds the writing across all styles and grammatical flairs. Our Sifter Factory produces new and updated Sifters weekly.
What is a Sifter?
A Sifter is a piece of software trained to read text and look for a specific concept. It reads novel language, learns from experience, and improves over time.
Our Sifters use the latest AI technologies
Our Sifters use machine learning (ML), natural language processing (NLP), and generative AI (GenAI).
What is machine learning?
What is natural language processing?
What is GenAI?
Generative artificial intelligence (GenAI) is artificial intelligence capable of generating text, images, or other data using generative models and often in response to prompts. Generative AI models learn the patterns and structure of their input training data and then generate new data that has similar characteristics.
Fun Facts about the Sifter Factory:
- Each Sifter is built in our Sifter Factory by a team of data scientists and attorneys equipped with the most advanced AI in the world to train algorithms that can read a contract.
- Each Sifter goes through a 9-stage development process involving 1-2 data scientists and 2-3 lawyers. We release new and improved Sifters every week to all of our clients at no additional cost.
- For the past several years, our Sifters come out of the Sifter Factory with 95-97% F1 scores. These scores continue to improve over time and with use.
Sifter Trainer
How it works
Our clients constantly contribute to the ongoing improvement of Sifters.
-
Our Sifters continuously improve with crowdsourced feedback from our clients, partners, and team members.
-
The Sifter Trainer is a feature of our software for users and our Sifting Services team to report Sifter errors.
-
Improved Sifters are released back into production weekly.
-
When we started in 2013, our Sifters were performing at F1 levels of around 90%. Now they exceed 97% thanks to technology advances and ongoing feedback from our users.
Benefits to you
-
Our clients are up and running immediately since they don't have to train any models before using our software.
- New clients get the benefit of all the learnings gained over our company's history and from our other clients, built into the Sifters themselves.
How we measure Sifter quality
-
We measure Sifter quality with an F1 score, a measure of predictive performance commonly used in statistical analysis of binary classification and information retrieval systems.
-
F1 combines both precision and recall into a single metric, providing a balanced assessment of a model’s effectiveness.
- Here’s how it’s calculated:
· Precision: The number of true positive results divided by the total number of samples predicted to be positive (including those not identified correctly).
· Recall: The number of true positive results divided by the total number of samples that should have been identified as positive.
The F1 score is the harmonic mean of precision and recall. It symmetrically represents both precision and recall in one metric. The value of the F1 score lies between 0 and 1, with 1 indicating perfect precision and recall, and 0 if either precision or recall is zero.
The F1 score balances the trade-off between precision (positive predictive value) and recall (sensitivity) and is commonly used to assess a model’s performance.
