Live Analytics Dashboard

---

• Date: April 2016 (First version, updated regularly)

Go to dashboard

First cloudsourced enterprise IT network analytics service.

This project started at Nyansa when we realized that we had access to a lot of useful information for network engineers. We decided to use this data to provide industry insights to everyone, allowing network teams to get answers to hard questions they could not answer on their own.

I led this project from idea, to prototype and finally production version. Using mostly RShiny with some custom JS bindings I was able to create a live analytics dashboard that highlights interesting features of enterprise networks.

Close

Social Dynamics on StackOverflow

---

• Date: December 2014

See report

This project was realized in a team of three for the Computational Social Science class at Stanford.

Using data dumps from StackOverflow (>80GB) we have studied the influence of getting the Legendary badge on the behavior of hardcore users. Based on our results it appears that after getting this badge (the hardest to get) users tend to answer less frequently. This suggests that there is indeed an incentive to answer in order to obtain such badges.

We also conducted descriptive analyses of the demographics of the users of StackOverflow.

Close

Job classification based on LinkedIn summaries

---

• Date: June 2015

See report

This project was realized in a team of two for the Deep Learning for NLP class at Stanford.

We used data provided by the company HiQ Labs that consisted of job title and job summaries as filled by users on LinkedIn. We reduced the number of output classes (job categories) using word vectors trained by Google and then matched these to job categories as defined in the ISCO-08 standard.

Once we had defined our output class for each job summary we trained a neural network to accurately predict the job category by reading the job summary as filled by the user.

This method reached an impressive accuracy of 31.7% to be compared with the baseline of 16% (predicting only the most common class).

Close

Stochastic viability of cities facing flooding risk

---

• Date: June 2013

See report

This document has been created following my research internship at the applied mathematics lab of Ecole des Ponts.

During my internship I developed a model of a city facing flooding risk. Decision makers want to protect the city by improving the drainage system and the seawall but they are constrained by economic resources. They also want to minimize the costs induced by floods. My model finds the best course of action using historical data about the floods and assesses the risk that the city faces.

After my internship I was offered a Teaching Assistant position for which I created this practical class. The aim was to illustrate the use of stochastic control on a real-life example and walk the students through a step-by-step Scilab model.

Close