Private Preview

The Market
Is A Graph.

Graph-derived signals and embeddings for systematic trading.

Drop-in features. Point-in-time safe. Built for research and production.

Request Backtest Artifacts

Private preview · Offline evaluation · No live trading required

64-dim Graph Embeddings300+ US EquitiesParquet + Schema

Graph Intelligence Preview

From Shock To Signal

Track how shocks propagate across connected assets, then inspect the ranked long/short signals produced from that structure.

Propagation Preview

Oil Shock Transmission

T + 3m
-5.92%-4.62%WTIJETSXLEDALUALAALXOM

Signals Feed

Top Graph-Derived Calls

Live ranking
Active Signal Thesis

Oil supply disruption (+8% WTI) propagates through airline cost structure via JETS ETF, creating a high-confidence SHORT opportunity in DAL.

WTI → JETS (−5.92%) → DAL (−4.62%)Exp Return -3.80%
Open full graph workspace

System Specifications

Engineering Superiority

Architecture
Heterogeneous Graph Models
Temporal GNNs optimized for financial causality. Natively fuses price, sector, and macro topologies.
Topology
Dynamic Multi-Edge
350+ Stocks, 10 Sector Indices, 14 Macro Indicators. Edges re-weighted by real-time correlation regimes.
Latency
Target: < 50ms
Designed for Apache Kafka ingestion and gRPC delivery. Optimized for high-frequency decision loops.
Infrastructure
Multi-GPU Clusters
Training and inference optimized for NVIDIA A100 clusters. Distributed graph processing for massive scale.
Features
Causal & Point-in-Time
80+ engineered signals including Fractals and Number-Theory Rhythms. Strictly leakage-safe.
Embeddings
Dynamic Vectors
64-dim node embeddings updated per minute. Encapsulates Lead-Lag and Shock Propagation mechanics.

Performance Delta

Information Gain

Metric (10-Month Walk-Forward)Baseline (XGBoost)RavenGraph FeaturesDelta
Implied Sharpe
Downstream Signal Quality Proxy
1.102.45+1.35
Predictive Lift (AUC)Baseline+15%+15%
Information Coefficient (IC)0.020.05+150%
Regime RobustnessLowHigh
* Results based on internal backtest (2023-2024). Past performance is not indicative of future results.
Research Integration

Built for
Feature Integration

Retrieve leakage-safe graph embeddings and join them directly into existing models.

Point-in-Time Embeddings

Embeddings are generated as-of a specific timestamp and horizon. No future data. No normalization leakage.

Model-Agnostic

Designed to plug into existing research pipelines (XGBoost, linear models, neural nets).

Offline First

Evaluate via Parquet artifacts before any live integration.

research_notebook.ipynb
1
from ravengraph import Client
2
import pandas as pd
3
4
rg = Client(api_key=API_KEY)
5
6# Point-in-time graph embeddings (leakage-safe)
7
embeddings = rg.embeddings.get(
8
universe="dynamic_sp500",
9
asof="2024-11-15",
10
horizon="1d",
11
dims=64
12
)
13
14# Join with existing feature matrix
15
X = price_features.join(embeddings, on="ticker")

Private Preview · Offline Evaluation · Parquet + Schema Included

Embeddings updated daily and intraday depending on horizon.