Engine 03Alpha Intelligence Engine
SYNATRA
Predictive Machine Learning Alpha
Synatra ingests structured market microstructure alongside alternative data — text, satellite, transactional — and extracts features through a stacked representation network. Signals are decay-modelled, capacity-tested and published directly into Qubex and Valtrius.
2.4k
Live signals in production
01 — Data Flow Architecture
Neural Signal Matrix
Raw inputs propagate through feature extraction layers. Pulse waves show activation flowing toward the final alpha signals. Hover a neuron to isolate its paths.
Live Simulation
- 01Alternative data ingestion
- 02Learned feature extraction
- 03Signal decay & capacity modelling
- 04Walk-forward validation harness
02 — Institutional Technical Specifications
Latency Profile
- Online inference
- 11 µs p50 (GPU)
- Feature refresh
- 250 µs – 1 day
- Signal publish
- < 2 µs to Qubex
Memory Footprint
- Feature store (hot)
- 1.8 TB in-memory
- Model ensemble
- 640 models
- Training corpus
- 9.4 PB
Data Coverage
- Market data
- Tick-level, 38 venues
- Alternative sets
- 120+ vendors
- History
- Since 1998
Audit Trail
- Model lineage
- Data → weights → signal
- Reproducibility
- Hash-pinned datasets
- Review
- Independent model risk
Enter the Synatra workspace.
Authenticated clients can open the live engine console: telemetry, configuration and audit trails in one workspace.