Skip to main content

How The Technology Actually Works - A Deep Dive

The User Journey: From Request to Result

Let's follow a specific example to understand exactly how this technology works. Imagine Ibu Sarah, a small business owner in Jakarta, wants to use AI to analyze her business finances and get a loan recommendation.

Step 1: Request Initiation​

The user opens her Mandala wallet and navigates to the AI services section. User selects "Financial Analysis and Loan Recommendation." The system already knows the user's verified business owner through Mandala-ID, but it doesn't need to know her name or specific identity for the AI to work.

Step 2: Data Preparation​

User’s financial data revenue, expenses, inventory, customer patterns exists in her Mandala wallet. When a user authorizes the AI analysis, this data doesn't get sent anywhere in readable form. Instead, it goes through a process called "secret sharing."

Think of it like this: Imagine a user's financial report is a painting. The system doesn't send the painting to anyone. Instead, it creates multiple overlapping transparencies where each one shows random dots and colors. No single transparency shows anything meaningful. But when you stack them all together in exactly the right way, you see the original painting. The mathematical version of this is far more complex, but the principle is the same.

Step 3: Distributed Computation​

These data fragments get distributed to different Nillion compute nodes. Let's say there are 7 nodes involved:

  • Node 1 in Singapore receives fragment A
  • Node 2 in Jakarta receives fragment B
  • Node 3 in Bali receives fragment C
  • And so on...

Each node receives what looks like random numbers. Node 1's data might be: [4829, -291, 8372, -4421...]. These numbers are meaningless without the other fragments. Even if a hacker compromised Node 1 completely, they would learn nothing about Ibu Sarah's finances.

Step 4: Blind Processing​

Each node performs mathematical operations on its fragments. They're essentially solving parts of an equation without knowing what the equation represents. The AI model for financial analysis has also been split into fragments across these nodes.

The nodes communicate with each other, but they only share computed results, never raw data. It's like a group of people working together to solve a jigsaw puzzle where each person can only see their own pieces and can only tell others things like "I have a blue edge piece" without showing the actual piece.

Step 5: Result Assembly​

After processing, each node has a fragment of the result. When combined using the appropriate mathematical protocol, these fragments reconstruct the final answer: "Loan approved for up to 500 million rupiah at 8% interest rate, based on strong cash flow and growth trajectory."

Step 6: Verification and Delivery​

The result comes with a cryptographic proof—a mathematical guarantee that:

  • The computation was performed correctly
  • The exact AI model specified was used
  • No data was exposed during processing
  • The result hasn't been tampered with

The user receives loan recommendation without any node, Mandala Chain, or even Nillion ever seeing her actual financial data.