Built and deployed a binary classifier that predicts whether a driver is going to accept or refuse a ride offer: It helped increase the number of rides as well as improve user experience. I also did a detailed analysis that showed the added value of such model before implementing the micro-service.

Built a first version of a marketplace simulator that allows the testing of new algorithms/ideas without having to deploy a new feature in production, and to go though the AB testing phase. I used state machines to model drivers and passengers behaviours. Transitions can be governed by existing machine learning models that are already in the stack to makes the simulation more realistic, it also allows testing potential changes in these models or to create new ones. The results of the simulation are charts showing the main KPIs of the business to have a better idea of should we invest our time into to improve a given KPI.

Built and deployed a binary classifier that predicts whether a client can pay the ride in the near future when payment authorization fails: we decide to give them a credit when the model is confident that the rider will pay the ride in time. It avoids missing out on would-be rides, improves user experience, builds trust and reduces fraud. I also did the integration of this in a very delicate place in the stack, lead the AB testing phase where this model proved to be better than previous systems.

Participated in writing the micro-service that performs weekly segmentation on the drivers fleet from a performance perspective: It helped us build a more effective incentives program and helped the Partner Relations to adapt their way of communicating with the drivers given the cluster they were in. It also gave a better understanding of the fleet in general.

Participated in building an in-house routing engine that uses historical data to provide accurate ETAs (estimated times of arrivals): It helped limit our costs while maintaining our user experience. It proved to be better than OSRM basic ETAs and it had an average error of 1 minute when compared to Google ETAs.

Participated in building the Kapten’s Machine Learning Architecture: - built a project template generator for Machine Learning micro-services that contains all the common elements that can exist in ML micro-services: the API, the database connector, the message broker client, etc. I participated in the the creation of the our Data Provider Gateway. This architecture helped increase the team’s productivity by spending less time on writing similar code in different places and to spend more time delivering value.

Wrote a micro-service that fed BigQuery tables by data coming in realtime in the message broker (rabbitMQ). It allowed us to get more data in our data-lake to address more problems such us "Driver Acceptance Prediction", "Ride Offer Screen AB Test".

Participated in writing my very first micro-service in GoLang. It aggregates data in realtime coming in the message broker and stores the results in a Mongo Database, it exposes an API to show different KPIS on the BackOffice, and in the App.

Built and deployed a micro-service prototype that switches rides between drivers during the approach to reduce ETAs and increase bother riders and drivers user experience: We used graph theory and applied a linear programming algorithm to solve this problem.

Proposed drivers incentives strategies to kick-start the business when the market was supply-constrained. It relied on the driver segmentation mentioned earlier. I built a simulation tool to prove the return on investment. And I also participated in the development of the micro-service and the monitoring.