Alain Gysi

The PPI-Doppelganger project was commissioned by Perreard Partners Investments (PPI) to explore the use of AI in financial markets. The initial goal was to build a deep learning model capable of predicting currency pair trends based on complex historical data to assist professional traders in making data-driven decisions.

I developed a predictive time-series model using TensorFlow and Keras, feeding it diverse financial indicators including previous OHLC (Open, High, Low, Close) data, inflation indices, and gold prices.

Explainable AI & Pivot to NLP

In the financial sector, a "black box" model is unusable. To ensure the model's predictions were transparent and trustworthy for PPI's traders, I heavily integrated Explainable AI (XAI) techniques using SHAP and LIME to map exactly which market factors were driving the model's output.

Despite rigorous optimization, the volatile nature of the market capped the predictive accuracy at around 55%, and the model required constant, time-consuming retraining to remain relevant.

The Strategic Pivot: Recognizing these limitations, I worked with the client to successfully pivot the project's focus toward Natural Language Processing (NLP). Instead of predicting raw market numbers, the architecture was redesigned to perform text classification, document summarization, topic modeling, and sentiment analysis. This new tool empowers PPI's traders to rapidly digest and extract critical insights from dense financial documents, bank reports, and global news articles.