Victor – Python, NumPy, Data Science

The project involved reimplementing a chatbot engine for a prominent bank in South America, transitioning to a Generative AI-based approach. The chatbot served as a knowledge retrieval tool for branch managers. Alongside the new engine, an updated operating model was devised for the relevant departments, encompassing a reevaluation of their routines, processes, and team structures. With approximately 60,000 users, the chatbot contributed to a notable increase of 40 percentage points in NPS (Net Promoter Score), while simultaneously achieving a 30% reduction in operational costs.

Victor has successfully achieved the following:

• developed methodology to evaluate answer correctness automatically;
• led a team of 3 data scientists to implement components of evaluation;
• coordinated business areas to evaluate and curate the chatbot knowledge base;
• reduced the operation costs by 30%.

The developed system implemented a fraud detection solution tailored for a Vehicle Insurance Company. Utilizing an Artificial Intelligence algorithm, claims were systematically classified and prioritized to optimize money recovery efforts. The implemented model achieved a remarkable six-fold increase in recovered funds compared to the client's previous system.

Victor has successfully carried out the following responsibilities:

• Created a fraud detection system and led the development of the algorithms;
• Carried out knowledge handover with the client's technical team;
• Improved the fraud area revision process;
• Developed methodology for testing new algorithms;
• Deployed model through an API using Docker, where the other systems could consult the fraud probability of a claim.

The project focused on the Post Merge Integration (PMI) of a CPG Company, aiming to optimize the new footprint to reduce production costs. An optimization model was devised, resulting in a significant cost reduction of $1 million annually.

Among others, Victor managed to achieve the following:

• developed ETL process and all the pipelines;
• developed optimization model;
• conducted meetings to translate technical knowledge;
• planned work and prioritized tasks;
• created several scenarios to simulate distinct operation changes, in order to recommend alternatives to the client.

The project involved implementing a churn detection algorithm for one of the largest banks in South America. This algorithm accurately predicted customers who were more likely to deactivate their accounts. A complementary model was developed to suggest product enhancements aimed at retaining clients, such as increasing credit card limits, lowering loan interest rates, or upgrading credit card categories.

The implemented system resulted in an annual revenue generation of $200,000 and a notable decrease in the churn rate by 12 percentage points.

As a part of the responsibilities, Victor performed the following duties:

• gathered all the necessary data for the algorithm features, model validation and performance estimations;
• carried out the updates with manager-level stakeholders;
• developed churn detection algorithm;
• developed product recommendation algorithm.