The problem Darth VDR - Virtual Data Room solves

1. Data Room is used extensively in the corporate sector when two entities want to discuss sensitive financial data. Both parties bring their proprietary data and make sure there is no copying of files or leakage of proprietary information to the opposite party.

2. Darth VDR is a Virtual Data Room in the Account Aggregator Framework. The FIU brings its algorithm and can run it in a secured isolated environment on the FIP data. No sensitive information is acquired by the other party and no information is stored in Darth VDR.

3. It is hard to run proprietary algorithms in a Data Room even if the data is digital due to security and trust issues. Portable devices cannot run computationally-expensive algorithms like ML models. Darth VDR can be scaled to support such use-cases.

4. Proposal:
   i. Virtual Data Room hosted inside AA - Most trusted entity in the entire AA Framework.
   iii. RBI regulations prevent AA to access the data - and these regulations hold in the proposed Virtual Data Room.
   iii. New consent type for VIRTUAL_DATA_ROOM should be added along with STORE, QUERY, STREAM and VIEW.

5. Process:
   i. The FIU sends a zip file with their proprietary algorithm to Darth VDR.
   ii. Proprietary algorithms can only fetch data from an allowed list of providers.
   iii. Darth VDR can only send out responses in an approved format like booleans or a predefined list of items.
   iv. No Personal Identifiable Information (PII) is sent by the VDR.
   v. Once an algorithm is saved, it can be used multiple times on different sets of data.

Functions

1. FIU submit code as Functions. Our solution supports AWS Lambda spec currently.
2. Valid JsonSchema for the Response validation.
3. Reusable by the FIU.

Jobs
Job is a request to process data using a Function submitted by FIU.
FIU schedule a Job after consent is received from the customer.

Command Line Tool

Challenges we ran into

• The Virtual Data Room topic was very open-ended and it took some time to narrow down on the approach.
• Full Homomorphic Encryption was a dead-end - We were able to get Paillier cryptosystem working.
• Lots of moving parts in the system for a small project.
• Constantly changing schema caused a lot of churn on both the Frontend and the Backend.
• The team is located in different time zones, so that has its issues.