This was a very challenging task, to profile the populations in each region using the aggregatd public info of each election results. We had long discussions on the model to use, but once we understood which was the Graphical Model to use, we did an E-M inference and obtained the latent variables. We exploited this information to the maximum, and one of the most important products was the Electoral Fertility Map. From this, our client maximized its results by targeting its efforts exactly in the most fertile regions to harvest just what it was looking for: votes.

IArxiv.org is a web platform designed for scientists. It learns the user topic preferences using LDA-algorithm (Topics Model) and then sorts each daily batch of papers accordingly. Represents a very important assistance for scientists daily life.

Dengue epidemic is a complex two-point function, since it needs two mosquito bites at different times to spread the epidemic. There observed variables (filled circles in the plot) and latent unobserved variables (empty circles). Bayesian Machine Learning allows to use prior information on each variable, and learn the relationship of the variables as well as the value of their magnitudes. Reaching at the end a probability distribution for the infected mosquitoes at each block (H3 hexagon), which provides an early estimation of dengue outbreaks, where Governments can apply early public health policies to prevent the outbreak.

We used data-analytics to detect fraud in physicians interaction with patients. Among other techniques, we computed the probability distributions of patients in different variables and then tested which physicians had their patients less likely to be sampled from the mentioned distributions. We studied these few cases and find many interesting things. A finding that the Pharma highly appreciated.

Gender-based violence is a profound problem in many societies. The National Congress asked us to develop a Smart Board that can track, classify and display, in order to raise awareness in society about the aggression that congresswomen constantly receive because of their gender. This board constantly scrapes social networks and measures in an intelligent way vectorizing a database from understanding the expressions in the treatment of female representatives. (see demo)

We are main AI service providers for IntelSocial (@intelsocialbot in Telegram, give it a try). We continuously scrap all Argentinian media (1000 newspapers, continuously) and relevant Social Networks, and we perform different level and depth AI analyses on the data to produce new tools and indicators. For example, we can detect when media with the same editorial line mono-editorialize and try to establish their news. We can also make heat-maps with all the media in the interior of the country/region detecting geographic patterns of specific news; a map with a pattern in some region may indicate excesses due to some specific reason, for example insecurity, or traffic accidents. A tool that increases the power and efficiency of public policies.

We designed a new 9-lecture course for the Executive Leaders at JP Morgan (VP and above). The objective was to learn the main algorithms, tricks, purposes, scope, true success and fail possibilities, and intuition on data and algorithms within the AI and Machine Learning disciplines, but for Executive that should lead teams and take decisions. The objective in each customized lecture is to understand AI without the need of seeing nor writing one line of code. This was a very successful coaching and we repeated to the Argentine high level executives, and a few verylarge companies within the country. We are very proud of it, because we have fully designed it and it worked out very well! Participants take a lot of profit not only from our knowledge, but also from our experience as practitioners who have made many working apps and tools.

Socio-Economic map for Buenos Aires. We've realized that the answers to all census questions (79 at each home) are not randomized at all, but instead they have a confounding latent variable. Since all questions are mainly socio-economic related, we argue that this confounder should be the socio-economic level. Henceforth, we've used matrix factorization techniques to infer this latent variable and we've obtained the first continuous heat map for socio-economic level for Argentinian cities. This index is continuous, and not discretized, and unsupervisedly obtained, which provides a crucial leverage when using it in many analysis.

Using Data from 911 phone calls and confirmed cases, we learned the pattern distribution from the Data and construct a software that could estimate number of COVID cases by analyzing number and origin of 911 calls. Including a scientific article published in Royal Society.

We have developed our own voice-to-text system in Spanish and we run it in our own servers. We are currently using it at IntelSocial (@intelsocialbot, test it as "/radios"), where it has a multi-purpose use in radio and TV streamings. For instance, to verify the correctness of advertising according to standards, or non-traditional advertising, or to analyze news items about certain topics (people, companies, etc.).

Doing basic research and teaching AI courses keeps us constantly at the state-of-the-art of the discipline. It forces us to keep learning new techniques, which we then use in our Apps & Tools developments. We also organize Workshops in specific subjects on which we want to expand our knowledge. We were born in the academy, we are from the university. Here some of our papers using AI for basic research: link-paper,link-paper,link-paper,link-paper,link-paper,link-paper, etc.

We were asked by a consortium to develop a method with Artificial Intelligence to predict variations in cryptocurrency prices. Something agile, in High Frequency Trading, at the minute. We used LSTM networks, and most importantly we set the error and uncertainty bars, and came up with a very interesting result. We could predict rises and falls to more than 3-sigma if these rises and falls were less than the Binance-fee. On the other hand, when the variation exceeded the fee, the error bars grew rapidly and any conclusions were diluted. It makes sense! We coined it the Market Uncertainty Principle. The most interesting thing about this work is that the answer we gave to the consortium was "there is no solution", at least at this level of sophistication. And they always congratulate us, as from this negative result they got out of cryptos.... at its peak!

Intelligent Parliament consists in an AI-analysis for the Congreeemen and Congresswoomen speeches using Topics-Model techniques. We used an unsupervised AI model to understand 20+ years of speeches in terms of their contents and construct a front-end to allow citizens to visualize and better how their representatives express their ideas.

Many organizations are willing to implement Machine Learning techniques, but they don't where to start. This is the optimal starting point. We sign NDAs and we dive into the company's data. We understand its acquisition, the data-flow, how it is used, and how it is not used. After usually a few weeks, we can write a detailed report that states not only many interesting AI ideas, but also we justify and suggest modifications in the data acquisition, data-flow, etc. We propose specific AI algorithm implementations that could be addressed. We provide a professional estimation on which implementations are likely to work and which not, based in our experience, skills and know how. We determine the enhancement of each proposed implementation. We estimate resources (hardware, software AI and developers hours) needed for each proposal. This report is an asset itself for the Company or Institution, since provides them a professional assessment customized for their organization.

Qlink.it is a web implementation of Quantum Criptography algorithm (only the algorithm!). It can be used to send secure messages through any plattform. (Video)

AI.Friendly.com is an open educative plattform in which users can easily understand and train a Neural network for the first time.

AI tool developed to learn mobility patterns and predict next day mobility index. Its main utility is to fast detect anomalous mobility. Developed together with Telefonica, one of the largest communication companies in the World

Mobility Index was designed as a tool for National and Province Governments to calibrate the lock-down. Thanks to this tool Governments designed activity spread to reduce contagion while re-activating economy. Work done in collaboration with Telefonica, one of the largest communication companies in the World.