Clients

Request a Demo

Use Cases

read the case

read the case

read the case

read the case

Client

Europa Press is the leading Spanish private news agency

Use Case

Explainable AI for combating fake news

Industry

Media

Our Role

To develop, alongside Europa Press, the fake news algorithm, using advanced models of Deep Learning, NLP and Explainable AI to understand how and why the model has made the decision. The result is presented in a dashboard for journalists with a reliability rating.

The solution is deployed on Google Cloud.

Key

AyGLOO’s ability to intuitively provide interpretability to AI models in order for them to be used by journalists without knowledge of data science.

Situation

A prediction by the consultancy firm Garnet in 2017 focussed on 2022 and claimed that, by this year, the western audience would consume more false news than true news. There is no way of checking whether we have already crossed that threshold, but concern about the rise of fake news is now at the centre of public debate.

For its part, the Ministry of Foreign Affairs, European Union and Cooperation, deems that lies and misinformation represent “a global threat to freedoms and for democracy” that has worsened due to social networks and the fact that “in recent years the flow of information and disinformation has accelerated”, as shown with the COVID pandemic. A reality that worsens when you take into account studies that point to the fact that eight out of every ten Spaniards have difficulties distinguishing between false and true news.

Solution

The Smart Fake News Detection (FND) tool is characterised for linking various different artificial intelligence techniques with regard to the news text subject to analysis such as NLP (Natural Language Processing), STS (Semantic Textual Similarity), NLI (Natural Language Inference), Detection of Anomalies and XAI (Explainable Artificial Intelligence). The result of the tool is finally presented in a dashboard that alerts the journalist to identify fake news, including the reasons why the system has deemed the news to be false or true.

To detect whether a news item is fake or not, each news text goes through a sequence of artificial intelligence processes with deep learning that treat the text in a cascade system of steps (categorisation and labelling, semantic vectorisation / embedding, searching for similarities, inference, classification), which finally determines its reliability.

All of these processes also follow a parallel explainable artificial intelligence (XAI) processing in order to maintain transparency and trust in the process followed by our FND tool.

return to cases

Client

Visual URB is a Spanish company that is digitalising urban planning with the aim of providing citizens and professionals from the sector with up-to-date urban planning data.

They group, classify and narrow down all of the urbanistic information and present it in a simple and updated way.

Use Case

Automation of the day-to-day extraction process of official bulletins for urban planning regulatory changes in Spain.

Industry

Digitally Native – Services Company

Our Role

To develop a tool to automate the extraction of urban planning data from all of the bulletins of the autonomous communities and provincial councils.

Key

The capacity and years of experience of the team at AyGLOO to apply NLP techniques we developed ourselves and deploy them serverless on the Cloud with a very high precision.

Situation

The client needs to have updated all of the information on urban planning in Spain and, until now, it has been a labour-intensive manual process that requires people searching for the relevant information in official bulletins every day. It is an inefficient process that consumes a lot of time and resources to do it properly.

Solution

The daily scraping system of the 62 websites of the autonomous communities and provincial councils to download pdf bulletins and extract all of the urban planning adverts, semantically interpreting the text by applying NLP techniques and identifying 19 fields in each advert that range from the land registry reference, procedure, various dates, planning, etc.

The system that has been developed is deployed on Google Cloud with daily processing using serverless and parallel techniques. It is capable of simultaneously processing an undefined number of documents, obtaining the result in just a few minutes and at a highly optimised cost.

The result is delivered to the client in a JSON that the client integrated into their systems and is presented in a dashboard.

Before:

Daily processing time of bulletins: 31 hours.

Now:

Daily processing time of bulletins: 5 minutes of processing bulletins + 1 hour maximum for reviewing them.

Before:

People working on the process: 3.5 people per year.

Now:

People working on the process each year: 0.1 people per year.

return to cases

Client

One of the Big Four

Use Case

Regulatory radar to automate the daily process of searching for regulations in the approval phase in official bodies and social networks on a European, national, regional and provincial level.

Industry

Services

Our Role

To develop regulatory radar models using our own NLP techniques to present regulations in the approval phase on a European, national regional and provincial level.

Key

The capability and years of experience of the team at AyGLOO to apply NLP techniques we developed ourselves and deploy them in a serverless way on the Cloud with a very high precision.

Situation

The regulatory search at an approval stage is a labour-intensive process that currently requires staff from the firm to spend a great deal of time on performing tasks of little added value.

Solution

A scraping is performed of the websites belonging to official bodies of the EU, Spanish government, autonomous and local governments, national institutions and entities on a European level.

The relevant information is downloaded and our own NLP techniques are used to perform the scraping, select information, crop, filter, add and classify it be sectors. The result is presented in a dashboard in which each user can view the relevant information to them.

The user also has the option of filtering the information at the top of the dashboard and the option of setting alerts to be communicated by email.

The system is multilingual.

return to cases

Client

Use Case

Deep Learning to detect the lack of oxygen saturation in the blood using a mobile device.

Industry

Healthcare

Our Role

AyGLOO has developed the Artificial Intelligence model that detects blood saturation through mobile devices.

Key

AyGLOO’s capability and experience with partners that have spent more than 20 years working with Machine Learning and Deep Learning in the Healthcare sector.

Situation

Shortness of breath, or dyspnoea, is an extremely common sensation in our day-to-day lives and most of the time it isn’t related to a health problem that requires control or monitoring. However, in some cases it is a direct indicator of a disease and needs to be assessed and treated by health professionals, because its consequences can be fatal. More than 400 million people worldwide diagnosed with respiratory diseases like COPD or asthma are extremely familiar with this feeling. And during the first wave of the pandemic, more than 95 million people diagnosed with COVID19 had this problem.

Solution

Smart Dyspnoea is an early detection system of oxygen saturation through the voice using Deep Learning algorithms that enable monitoring to be carried out, and the degree of dyspnoea to be identified thanks to the application of a clinically validated test with 91% reliability (Roth Test).

The solution has been recognised for its social and ethical impact, among others.

return to cases

Client

Soni2 is a recording studio

Use Case

AI for cloning the human voice

Industry

Media

Our Role

To develop, alongside Soni2 with the collaboration of the University of Valencia, the voice cloning algorithm for use in the area of e-learning and audiobooks, using artificial intelligence models.

Key

AyGLOO’s ability to work successfully on an R+D project, applying artificial intelligence.

Situation

Human voice cloning has made progress in recent years, particularly in English. The state of the art in other languages, including Spanish, is still at an early stage.

Solution

The aim is to reproduce a speaker’s voice with their tone, pitch, accent and breathing rhythm. Using a text, acoustic and generative AI models that incorporate volume, pronunciation, intonation, speed and emotion to be able to reproduced the speaker’s voice. The system is trained with hours recorded of the speaker.

return to cases