RavenPack unveils AI-driven pipeline that creates backtest-ready financial datasets in under an hour

RavenPack launched a dataset-generation pipeline on its Bigdata.com platform that leverages a customized NVIDIA Nemotron model to label massive volumes of financial text quickly. By first fine-tuning a 120B-parameter Nemotron 3 Super model and then distilling it into a 4B-parameter Nemotron 3 Nano model, the system labeled 845,220 excerpts in roughly 49 minutes using two NVIDIA H200 NVL GPUs. The labeled data were transformed into daily company-level signals and backtested over a ten-year span, delivering an information ratio of 1.18 and an annualised excess return of 8.63%, far outperforming a sentiment-only approach.

Cost reductions include a 63% drop for a full historical dataset and a 96% saving on subsequent runs. The pipeline is now available to Bigdata.com clients via a companion notebook that handles the entire workflow from hypothesis to backtest.