Synapse
Broadening Data Boundaries with Privacy-First Interoperability
Created on 4th November 2024
•
Synapse
Broadening Data Boundaries with Privacy-First Interoperability
The problem Synapse solves
We are expanding upon the existing AA framework to overcome the following limitations:
- Data sharing is sufficient but not necessary for data empowerment. Financial insights reside in non-financial data that may or may not be in standardised formats.
- Current framework limits the coverage of digital footprint that a user can leverage only to FIPs aligned with a industry wide common data schema.
- Moreover, some non-financial data custodians (eg: Telcos) cannot share sensitive data (eg: location tracing) even if a standardised format was aligned.
This calls for an alternative approach of federated learning and inference where the financial insights can still be leveraged while making sure raw data never leaves the environment.
Our solution can potentially transform AA from a Data-sharing-framework to a highly interoperable Intelligence-sharing-framework thus expanding the FIP ecosystem to include more diverse data custodians at a faster pace.
Challenges we ran into
One of the primary challenges was that the SahamatiNet and ReBIT APIs are strictly designed for data transfers. These APIs focus on enabling data sharing within the AA ecosystem, making it difficult to implement the remote model triggers required for our solution.
To address this, we simulated FIU FIP and AA actors and developed custom APIs based on modifications of ReBIT's existing contract specifications to handle:
- New consent type of "Consent to query data"
- New request parameters on triggering different model types with different runtime parameters and data filters
- Passthrough mechanism for model outputs to flow from FIP to AA to FIU
This allowed us to trigger model runs to compute insights across diverse data sources securely, without needing direct data transfers.
Tracks Applied (1)