The legaltech sector is undergoing a rapid transformation, and we are proud to be part of it. The most innovative technologies of Natural Language Processing (NLP) and Machine Learning can provide smarter and more efficient services for the practice of law.
Our team at Komorebi AI has developed tools for a major provider of information services for legal offices. Our tasks were to develop NLP tools to structure legal documents written in plain text. These tools were applied to a large corpus of unstructured court rulings and allowed to build a well structured database. On this database, we could implement smarter and more accurate queries by specific content. In addition, we used advanced ML techniques to improve their document retrieval engine, by learning and tuning the concept of relevance to each user’s experience and preferences.
Products
● Custom Name Entity Recognition
● Structuring court rulings
● Verdict analysis and prediction
● Smarter information retrieval
Satlink manufactures smart buoys for the fishing industry, which are able to report to a satellite how much fish is concentrated on the water below the buoy, and even discriminate between different species.
Komorebi has developed several products for Satlink, that have contributed to improve the information services provided by their platform.
● Prediction of trajectory for drifting buoys, combining oceanographic models for currents and historic data of buoys movements.
● Spatio-temporal echosounder signal analysis to discriminate fish species.
● Backend data server that aggregates oceanographic and weather data with download scripts and access APIs.
The added value from this technology is:
● Reduction of lost buoys due to collisions with the coast.
● Route optimization for fishing vessels reducing emissions, fuel and operation costs.
● Decrease in bycatches, thanks to a more accurate species determination.
● Improved information and decision support systems for the fishing vessels.
Omnicom is a large company specialized in allocating and monitoring investment in marketing campaigns. We developed various products for them.
One of them was a model based in Bayesian time series forecasting that was able to forecast weekly sales at each point for a large chain of fast food restaurants, taking into account external variables (weather, social events, economic indicators, etc. ) and marketing resource allocation across different channels. This allowed them to assess the impact of marketing decisions, and optimize ROI to achieve larger sales with a fixed advertising budget, deciding when and where to allocate the investment.
Our model predicted the next week sales with an average error below 5% of the real value.
We developed CTR prediction models on data from ad servers that contained more than 100M monthly interactions between users and ads. Employing the most recent ML techniques for sparse big data we were able to achieve a substantial improvement on their previous performance.
● High precision
● Interpretable models
● Quantify uncertainty
● Risk analysis
Although less than 0.04% of the total amount of transactions, fraud in credit card payments cause losses to the banking industry for over 1.8 billion EUR only in the Euro zone, and the correspond mostly to internet payments.
The team at Komorebi has built machine learning models for fraud detection in electronic payments, trained on more than 150M credit card payments. We created an algorithm capable of predicting in real time the probability that a given transaction was fraudulent. The project involved also a decision support system for the head of fraud analysis at the issuer bank in order to manage alerts, optimizing the expected utility for the bank. The system calculated the best possible action on a transaction (allow, alert, stop) taking into account the expected cost of all possible outcomes and the customer’s profile. This project was financed by the BBVA Foundation under its prestigious Leonardo Scholarships Program.
a3sec is a company specialized in ensuring the robust operation of large networks of electronic devices. With a network size of over a few hundred thousand devices it is impossible to have a human supervision of its status and operations. Among its customers, a3sec provides services to a large chain of supermarkets.
For this project, the team at Komorebi analyzed thousands of time series to provide an automatic monitoring system based in dynamic Bayesian models. This system was capable of detecting anomalous behaviour of single devices signaling threats and guaranteeing its safe operation. The system also had predictive capacity, which allows to take actions before devices enter into a critical operation regime.
The value added by this product to the end user is a reduction of time lost by stops due to critical events, detection of compromised devices and overall operation monitoring for maintenance suggesting preventive actions.
● Outliers detection
● Efficient and scalable solutions
● Predictive maintenance
The Instituto de Salud Carlos III is one of the main research institutions for public health in Spain. The team at Komorebi AI has participated in a project whose aim is to study the impact of physical activity on the risk of cardiovascular diseases, performing a rigorous statistical study with big data.
We analyzed a total of 2M clinical records, enriching the study with other public data coming from external sources, such as demographic data provided by the Instituto Nacional de Estadística (INE).
The main challenge of this project was to develop tools to process a large amount of data that have great value but low quality. Much of the relevant information in clinical reports was stored as plain text, with a large variability due to wide range of doctors and patients. We develop tools to homogenize and structure these health records using natural language processing, in order to extract the relevant features and improve the quality of the statistical study.
