There will be links to some examples of the work done in the projects or of the techniques discussed with some other applications. I can not add links to the projects' repositories since they are private repositories and contain information that is subject to a NDA.
- Detect fuel prices from images of our clients competitors in order to have more recent and accurate data to feed regression models for pricing intelligence.
- Increased clients engagement with the CRM by approximately 20% (measured by the average number of daily clicks on the modules affected by this change).
- Python
- matplotlib
- easyOCR
- pytesseract
- cv2
- Process that gives a price recommendation for a gasoline station in order to maximize earnings considering: price of competitors, traffic information, price demand elasticity, a forecast for the daily sales, etc.
- R ('tidyverse' with an emphasis on 'dplyr' and the base 'stats' package) for data wrangling and modeling regarding Linear Regression.
- Python (sci-kit learn and keras), for sales forecasting and as a bridge betweeen R and SQL for faster insertions to the database.
- RMarkdown and Excel for communicating results.
- SQL.
- Segmentation of the ~13,000 gasoline stations in Mexico into ~1,500 50 km. radius clusters in order to lower the number of API calls from data sources regarding fuel prices and traffic information to reduce costs to approximately 20% of the original costs and to 80% of the costs after using a heuristic approach to group the stations. The approach used was to use Hierarchical Clustering, this way the clusters are created iteratively grouping the nearest stations, and the algorithm is stopped when the in the next iteration a cluster passes the 50km radius.
- Group gasoline stations by: predicted sales, level of competition, income level of location, etc. in order to recommend clients the most profitable regions to open or acquire a station.
- R
- SQL