Design
Overview of the Silsilat Architecture
Silsilat is designed as a modular, agent-based decentralized finance network built on Hedera Hashgraph. It integrates real-world gold collateral, AI-driven valuation, automated liquidity provisioning, and policy-driven compliance into a single interoperable framework.
At its core, Silsilat has three layers:
Layer
Purpose
Core Components
Application Layer
User-facing interfaces for pawnshops, investors, and regulators.
Web dashboard, mobile app, APIs, and agent control panels.
Protocol Layer
Core logic and liquidity flow mechanisms that govern how assets move.
Tokenization engine, liquidity pool contracts, and policy registry.
Consensus & Data Layer
Immutable recordkeeping and traceability for all events and transactions.
Hedera Consensus Service (HCS), IPFS artifact storage, and Arize Phoenix observability.
Together, these layers create a trust fabric linking physical gold to digital liquidity.
Key Components
A. SAG Tokens (Secured Asset Gold)
Each pledged gold item is represented as a SAG Token, minted when the pawnshop records a loan on Silsilat.
Backed 1:1 by physical gold stored at the pawnshop.
Contains metadata such as purity, weight, appraised value, and policy compliance hash.
Used as the unit of transaction within the liquidity pool.
When the loan matures, the SAG token is burned, and the pawnshop regains ownership of the physical collateral.
B. Liquidity Pool
The Silsilat Liquidity Pool acts as an automated market maker (AMM) that instantly buys SAG tokens from pawnshops at a pre-defined policy rate.
Investors contribute capital into the pool and receive Liquidity Tokens (LQT) in return.
LQT holders earn a return from the profit spread on loan repayments.
On maturity, SAG tokens are repurchased by the pawnshop, the liquidity pool releases funds to investors, and the tokens are burned.
This ensures real-time liquidity, similar to decentralized exchanges but backed by tangible assets.
C. AI Evaluator Agents
The Gold Evaluator Agent determines the fair loan-to-value (LTV) ratio by combining:
Real-time gold price feeds (via MetalPriceAPI).
Currency conversion (via FastForex).
Policy constraints (via
policy.py).Local appraiser data and model predictions.
Each evaluation is:
Logged via Arize Phoenix for trace analysis and model observability.
Published as an artifact on IPFS.
Indexed on Hedera HCS with a
trace_idand policy reference.
This creates a verifiable AI audit trail for each loan evaluation.
D. Compliance & Policy Engine
Silsilat enforces automated compliance through a modular Policy Engine. Each jurisdiction or financial institution can deploy its own policy pack (e.g., Ar-Rahnu, AML, or Shariah).
Policies define:
Maximum LTV ratios.
Minimum haircut for jewelry/bar gold.
AML/KYC thresholds (e.g., > RM 25,000 flagging).
Required trace fields for auditability.
These policies are versioned, signed, and anchored on Hedera, enabling regulators to validate compliance post-facto or in real time.
E. Hedera Consensus Topics (HCS)
Each Silsilat Agent has three dedicated HCS Topics:
Topic Type
Function
Input Topic
Receives new loan or collateral events.
Output Topic
Publishes analysis results and LTV recommendations.
Override Topic
Allows authorized administrators to override an AI decision in exceptional cases.
The override flow ensures human-in-the-loop accountability while preserving decentralized audit trails.
F. IPFS Artifact Storage
Every agent trace, compliance report, and model output is packaged as an IPFS artifact containing:
Input and output data
Trace ID
Model version
Timestamp
Policy hash
Only the IPFS pointer (CID) is stored on Hedera, ensuring data privacy and scalability.
End-to-End Transaction Flow
Let’s look at how a typical Silsilat transaction unfolds:
Trust & Traceability Framework
Silsilat achieves verifiable trust through:
Cryptographic Anchoring: Each event hashed and anchored on Hedera Consensus Service.
Trace Observability: AI inferences monitored via Arize Phoenix.
Audit Artifacts: Immutable IPFS records with signed policy hashes.
Override Controls: Human-admin channels for governance and dispute resolution.
This ensures accountable automation combining the efficiency of agents with the oversight of regulators.
Integration Overview
Silsilat exposes a RESTful API and SDK that allows:
Pawnshops to submit and manage tokenized loans.
Liquidity providers to deposit, withdraw, and view pool metrics.
Regulators to query compliance proofs and audit trails.
Future versions will support GraphQL queries and MCP (Model Context Protocol) adapters for seamless integration with AI agents and Web3 dashboards.
Last updated