In May 2022, we created the Data Capabilities squad, which mission is to support members and employees in making BlaBlaCar a safe and reliable marketplace. We are operating as a more or less autonomous team composed by 6 engineers (software eng, data eng, data analyst and data engineer). Together we provide product teams with data analytics, dashboards and data driven recommendations to improve authentication and trust features, payment method, messaging system, ... We own a fraud detection engine and produce transversal referential for other data squads.

Mission:As a lead data scientist, I have 2 missions:- 70% time: contribute as a senior data scientist on complex problematics - 30% time: lead the data scientist chapter, which mission is to maximise the success likelihood of problems addressed by blablacar data scientists. Organise best practices sharing between data scientist, improve our ML stack, collaborate with external experts (academics or other companies).Main contribution:- train and deploy pricing algorithms for carpooling rides- designed a machine learning fairness checks to measure how fair our pricing algorithms are in regards of the gender biais we observe on the carpooling marketplace.- build a model deployment framework to take care of the administrative part of model deployment (API, deployment scripts, prediction logs, decision thresholds, interaction with data sources, ...). Data Scientists can now focus mostly on the business need, model, and impact measurement- Improve the pricing model transparency and communication with stakeholders.- Design scripts and methods to estimate models impact without the need to go for an A/B test in production.- Collaborate with researchers on a study to measure passenger price, proximity and duration elasticity.

Mission: Support the company by challenging decision processes, detect where machine learning algorithms could add value, and deploy algorithms to take fast, risk aware decisions. At BlaBlaCar the data scientist is in charge of the entire ML model lifecycle: defining the need, doing a POC (create a model and estimate impact), communicating with stakeholders, deploying and monitoring production. Data stack: Google Cloud Platform (BigQuery, ComputeEngine, CloudRun, AppEngine, PubSub), SQL, Python, Tableau, Jenkins, DockerMain contributions:CRM optimisation:- Create and deploy a member segmentation to support CRM decisions. Recency Frequency Monetary matri. Create retention KPI’s (churn rate, cruise rate, new cruise members ratio)- Churn prediction: define churn, identify main churn sources, build and deploy an algorithm to predict churn in advance, trigger an AB tested CRM campaign to prevent churn (we measured +10% in retention metrics)- Setup an A/B testing framework for CRM and SEA campaigns- Next trip prediction: predict travel need, to make sure drivers don’t forget to publish their ride and passengers should consider blablacar in the travel optionsFraud detection:- Build realtime algorithms based on user website behaviour.- Build algorithm to detect linked accounts (a customer using several blablacar accounts)- Communicate to stakeholders about precision and recallBus networks:- Benchmark existing solution, make or buy decision, - Build a bus line frequency optimisation tool (addressed by genetic algorithms)- Communicate with bus operations, to support them with our tool

I was mostly working as a marketing data analyst, in a team of 5 people. The main mission was to improve the marketing performances by promoting a data-driven culture, tools and process to the marketing managers. I also worked on macro analysis such as user segmentation and French mobility market analysis.Tools and languages: SQL, Tableau, Excel, PythonMain projects and analysis:- Improve marketing performance KPI. Define better KPI to support the new growth phase (from acquisition, to activation, to retention). Update the data stack to compute, update the report, train marketing managers on how to make decisions based on the new reports.- Revamp of the marketing tracking technical stack. - Setup tracking partners. Supervised the mobile tracking partner change (Adjust). Challenged our facebook PMD (Preferred Marketing Developer)- Improve and automate reporting: migrate from Excel to tableau- Build a blablacar member segmentation to understand who is using blablacar. Segment the user base based on NLP analysis of the mini-bios in members' profiles, build clusters, and define personas. This was used to challenge our a-priori and discover under-penetrated personas. Communicate the results to the branding team.- French mobility market analysis: segment french intercity axes to create homogeneous markets, join with INSEE data to enrich the analysis, join with mobility study to understand BlaBlaCar market penetration.-Quantitative and qualitative analysis on French members retention issues.

Created reports (IBM Cognos):- invoicing and shipments following, for customers- invoicing performances reports, for managers- macro performance of the agency compared to forecast, for central decision makersData quality policy:- automated data quality checks- promoted data quality internally and teached agents how to properly tag shipments events

Airplane are not consuming exactly what was pretended by manufacturers. A coefficient is updated to take into account plane aging, defaults and consumption biais, the objective being to forecast consumption as precisely as possible.- History analysis to show project potential in terms of security and fuel saving opportunity.- Built a model to forecast next 15 days value for the coefficient (Summer Winter model, Kahlman filter, ...)- Quickoff of the implementation of coefficient forecast and updates in the "old" company and manufacturers tools

Among other ad-hoc missions, the main objective of the internship was to help the team in task automation:Every week, the team of 3 engineers was doing computation in each of the manufacturer tools, and compare the results with expected truth results. Once validated over software versions, runaways and aircrafts models , the configurations could be used in production (uploaded in the plane and dispatch), to serve take off and landing distances computations. This is critical to ensure security and it belongs to airlines to validate the tools.We decided to automate this process, to avoid this no added-value wasted time. - Create test case scenarios- Create a bot which is going through the user interface (language used: AutoIT, close to Basic)- Manage errors and restart on failure- Aggregate results on a dashboard (excel)

Audit of the production units performance tracking (Detected tracking failures leading to wrong decisions).Connected to databases to extract data and analyse in Excel.Built a macro month performance recap on production units.