Published 1 hour ago • loading... • Updated 1 hour ago

RavenPack launches AI-powered dataset pipeline

RavenPack launched a dataset-generation pipeline on Bigdata that transforms plain-language research hypotheses into investment-ready datasets using fine-tuned NVIDIA Nemotron models.
Data science teams struggle to convert research ideas into datasets with sufficient coverage. RavenPack fine-tuned an NVIDIA Nemotron Nano model using results from a larger Super version through LoRA supervised fine-tuning.
Running on two NVIDIA H200 NVL GPUs, the system labelled 845,220 passages in around 49 minutes, achieving around 39 times the throughput of the larger teacher model in RavenPack testing.
The resulting signal yielded an 8.63 per cent annualized excess return versus 3.78 per cent for sentiment-only approaches. Labeling costs dropped 63 per cent, with subsequent runs costing 96 per cent less.
NVIDIA Nemotron's open-weight architecture enables regulated institutions to deploy and audit the model internally. The pipeline is available immediately to Bigdata clients for complete workflows from hypothesis to backtest.

Insights by Ground AI

46 Articles

Chad

Reposted by

rotherhamadvertiser.co.uk

Center

RavenPack unveils AI-powered dataset pipeline to test investment ideas

RavenPack has unveiled a new dataset-generation pipeline on Bigdata.com that uses a fine-tuned NVIDIA Nemotron model to label hundreds of thousands of financial text excerpts in under an hour, helping investment firms test research hypotheses faster and at lower cost.

1 hour ago·United Kingdom

Read Full Article