AI Investments - Applied R&D for stable financial time-series forecasting

OVERVIEW

The project was delivered as a joint R&D effort with AI Investments, an established fintech company operating in Europe and the United States. The goal was to strengthen investment algorithms by improving the accuracy and stability of financial time-series predictions across a broad range of instruments. The research phase included structured evaluation of multiple forecasting techniques before converging on a final ensemble-based approach.

The work focused on adapting advanced machine learning architectures to the specific characteristics of financial data. This included extensive experimentation with Transformer-based models such as Temporal Fusion Transformer, Conformer, and Fedformer, adapted and validated for financial time-series prediction.

To support research at scale, 7bulls designed and operated a dedicated, multi-cloud machine learning platform. It enabled cost-aware experimentation and continuous retraining, supporting an average of approximately 3,000 models whose predictions were ensembled at one-hour intervals, without compromising operational efficiency.

THE CHALLENGE

7bulls combined applied machine learning expertise with long-term experience in financial-market R&D delivery. The team’s background in building and operating research-grade ML platforms enabled rapid experimentation, objective comparison of approaches, and a controlled transition from research to production.

A key differentiator was the design of automation-first research workflows, allowing thousands of models to be trained, evaluated, and retrained continuously without excessive infrastructure costs. This reduced research risk and ensured that outcomes were measurable and repeatable rather than isolated experiments.

By pairing technical depth with experience in managing complex R&D initiatives, 7bulls acted as a reliable partner in advancing a data-intensive fintech product from research hypotheses to stable production performance.