Embedded Files
Objective
Objective
Show that domain-constrained causal modeling (with attention as a mask) produces explainable, audit-ready signals that are more reliable than raw correlation and traditional blackbox systems. [View MVP specs.]
Constraint
Constraint
We had no premium exchange feeds and free public APIs were aggregated/rounded and hence unreliable. To test the idea fairly, we needed clean, authoritative data.
Data Access
Data Access
We filed RTIs to obtain regulatory datasets that are normally unavailable to institutions. All data was used strictly for research and to prove the concept only. No Regulatory data was/will be used for commercialization.
The PoC
The PoC
The PoC produced non-trivial, reviewable signals with contributor explanations rather than opaque weights.
Our model proves that in markets, precursors matter: quantifying what tends to precede an event improves our ability to anticipate the consequences that follow. With a hybrid Algorithmic ML + Attention model, we can defend those calls with reasons thus staying compliant to Regulatory standards.
Our gratitude to AWS for their startup grants—fueling the compute that allowed us to prove what we believed was possible.
"This is a GPU driven business—
Silicon is the new Gold.
Data is the new Silver.
Electricity is the new Bronze.
So we can eat Sapphires for breakfast."
- Shubham Sood (Founder)
Google Sites
Report abuse