At the intersection of mathematics, computer science and business, data science was a term that was coined to describe the techniques meant to extract meaning from data and creating data-scentric products. My curriculum brought me to data science through pure mathematics. Where there's quantity, uncertainty, rationality, or structure, there's something to be done with math. My job was to uncover those aspects, express the problem and objectives of the solution, and go about exploiting whatever mathematical aspects can be exploited in order to provide better solutions. This involves capturing and utilizing knowledge that is usually locked-up in clients' experience and/or data.

I worked as a mathematics consultant since year 2000. Generally, I used a mathematical approach to define, analyze, solve and/or optimize given aspects of my clients’ field. In particular, I developed new technological designs, audited and optimized business processes, and elaborated automatic forecasting and targeting solutions for marketing problems, mining and utilizing my clients' knowledge of the field, and whatever data they were wise enough to have collected.

In my first years of consulting, I have applied myself to a wide range of disparate problems. More recently, a few recurring themes have emerged from many of my projects. One of them is that of developing automated systems that can carry out a certain level of behavioral analysis. Whether assessing the tastes of clients, modeling the spread (or not) of viral campaigns, analyzing the marketing strategies of companies through the emails they send or ads they disseminate on the web, modeling the behavior of drug use, or predicting the evolution of airfares; these all involve a certain level of analysis of the behavior of the agents (individuals, social groups, companies, automatic systems, etc.) involved.

You may get an idea of my most recent activities on the PROJECTS page.

I have a predilection for modern techniques that take advantage of our era's computing power, such as Bayesian methods, optimization meta-heuristics, graphical models, etc. Other methods/domains encountered regularly include: Cluster analysis, neural networks, information retrieval, machine learning, Markov chains, stochastic analysis, social network analysis, recommender systems, collaborative filtering, etc. The final result usually involves a mix of these approaches and good amount of home grown techniques, inspired by the specific problem at hand.

Generally, a consulting project will involve a work flow resembling the following:
  • Talk with and learn some elements of the expertise of the client.

  • Carefully define these elements and the objectives of the client.

  • Quantify and structure the relation between these elements.

  • Search the literature for previous applications of mathematics to the domain.

  • Train the client's engineers or help the client find the right data scientist profiles to hire.

  • Develop innovative methods, formulas, algorithms, and/or full solutions from a combination of the client’s expertise, state-of-art research, and novel self-grown techniques.