Welcome to Phase 2 of Project Persephone, our showcase for rapid and cost-effective Stock Volatility Prediction in a strategic partnership with AWS.
We developed a complex, dual-model system to forecast market behaviour: a 'sentry' model for calm periods and a 'hunter' model for stormy ones.
Incredible Efficiency: Proof of Concept Results
This was achieved using an enterprise-grade MLOps pipeline and a high-performance cloud architecture.
The High-Performance Tech Stack:
Compute: AWS p4d instance (8x NVIDIA A100s, 96 cores, 1TB RAM)
GPU Acceleration: NVIDIA CUDA, RAPIDS, and Dask for distributed computing.
ML Models: LightGBM and XGBoost.
MLOps: A full pipeline with TensorBoard for observability and accuracy tracking.
Storage: A tiered strategy with NVMe (scratch), EBS (persistent), and EFS (dataset).
Networking: 400Gbps with Elastic Fabric Adapter
The next stage for this project is to enrich the models by scraping and analysing financial news. This project proves that high-impact financial AI is achievable with incredible speed and efficiency.
OUR PARTNERS:
