Architecture Details

Core Architectural Principle

Silsilat is built around the principle of Accountable Automation — every decision made by an AI or agent is observable, auditable, and anchored to a verifiable event on-chain.

The architecture unites AI, DLT, and liquidity protocols through three core primitives:

1. Agents — autonomous evaluators for valuation, compliance, and liquidity actions.

2. Topics — Hedera Consensus Service (HCS) channels that record immutable event logs.

3. Artifacts — IPFS-stored bundles that preserve full trace evidence and model context.

Together, these form a verifiable trust loop:

Event → Evaluation → Artifact → Consensus → Action.

---

High-Level Architecture Diagram

---

Module Breakdown

A. Agent Layer

Agents are independent Python services, each running within its own container or smart node environment. They communicate through HCS messages and external APIs.

Agent	Function	Data Source	Output
Gold Evaluator Agent	Calculates loan-to-value (LTV) ratio and haircut	MetalPriceAPI, FastForex, Policy registry	JSON payload with valuation and trace ID
Policy Agent	Applies KYC, AML, and Shariah rules	AML/KYC database, Regulatory schemas	Compliance pass/fail + reason codes
Liquidity Agent	Executes pool operations and manages token flows	Liquidity pool smart node	Transaction confirmation, pool metrics
Audit Agent	Monitors all traces and publishes integrity proofs	IPFS, Arize Phoenix	Hash proof and verification event
Override Agent	Allows human intervention on specific decisions	Admin dashboard	Updated decision record on HCS

Each agent uses OpenTelemetry tracing and sends observability data to Arize Phoenix for model performance monitoring.

---

B. API Layer

Silsilat exposes a secure RESTful API built with FastAPI (Python) and connected via Hedera SDK clients.

Example endpoints:

# Example: Submit gold loan
POST /api/v1/loans
{
  "pawnshop_id": "ARHNU123",
  "item_type": "gold_bar",
  "weight_grams": 15.2,
  "purity": 0.916,
  "appraised_value_myr": 4500,
  "borrower_id": "NRICxxxx",
  "policy_id": "AR_RAHNU_STD_2025"
}

# Example: Get LTV analysis
GET /api/v1/evaluations/{trace_id}

# Example: Publish to HCS
POST /api/v1/hcs/publish
{
  "topic_id": "0.0.556789",
  "message": {
    "trace_id": "uuid",
    "ipfs_cid": "bafybeihash...",
    "event_type": "loan_evaluation"
  }
}

Each endpoint enforces:

• JWT-based authentication

• Role-based access (pawnshop / regulator / investor)

• Rate limits and HMAC-signed payloads

---

C. HCS Topic Management

Each agent subscribes to specific Hedera Consensus Topics, as configured in the .env.local file.

Example configuration:

HCS_INPUT_TOPIC_ID=0.0.1001
HCS_OUTPUT_TOPIC_ID=0.0.1002
HCS_OVERRIDE_TOPIC_ID=0.0.1003

Message structure (JSON):

{
  "event": "loan_evaluation_complete",
  "trace_id": "7f1a9c9b-1c32-4c0d-a510-abc123",
  "ipfs_cid": "bafybeigoldtracehash",
  "policy_hash": "sha256:xyz...",
  "timestamp": "2025-10-26T04:35:00Z"
}

All messages are signed by the sending node and can be verified via Hedera SDK’s public key validation.

---

D. IPFS Artifact Schema

Each AI inference or policy validation generates an artifact bundle stored in IPFS:

{
  "trace_id": "7f1a9c9b-1c32-4c0d-a510-abc123",
  "model_name": "gold_evaluator_v2",
  "inputs": {
    "gold_price_usd": 2350.12,
    "fx_rate_usd_myr": 4.73,
    "weight_grams": 15.2,
    "purity": 0.916
  },
  "outputs": {
    "ltv_ratio": 0.78,
    "haircut": 0.12,
    "loan_value_myr": 5300.55
  },
  "policy_enforced": "AR_RAHNU_STD_2025",
  "timestamp": "2025-10-26T04:35:00Z",
  "phoenix_trace_id": "8663ccf15d0b5483..."
}

Only the IPFS CID and hash are recorded on Hedera; raw data can be retrieved by authorized actors using the CID reference.

---

E. Liquidity Module

The Liquidity Agent and Liquidity Pool Smart Node manage tokenized asset flow.

Lifecycle:

1. Pawnshop mints a SAG token.

2. Liquidity Pool automatically buys SAG token for stablecoin (e.g., USDC, Hbar, or MYR-t).

3. Investors deposit liquidity and receive LQT tokens representing pool share.

4. On maturity, SAG is redeemed and burned.

5. LQT holders receive yield and can withdraw.

---

🛡️ 4️⃣ Security and Compliance Controls

Control Layer	Mechanism
Identity & Authentication	DID-based identity registry (HCS-11), JWT tokens
Data Integrity	SHA-256 hashing of all artifacts
Audit & Traceability	IPFS artifact logs, Phoenix trace replay
Regulatory Reporting	Policy-based event filters pushed to Regulator Topic
Shariah Compliance	Rule-pack enforcement and structured audit evidence

---

🔄 5️⃣ End-to-End Event Lifecycle

Stage	Action	Ledger Record
1. Loan Created	Pawnshop submits item & metadata	HCS Input
2. Evaluation	Gold Evaluator computes LTV	HCS Output + IPFS Artifact
3. Compliance Check	Policy Agent validates AML/KYC	Regulator Topic
4. Liquidity Execution	Pool purchases SAG	Pool Transaction Record
5. Audit & Override	Trace verified or amended	Override Topic
6. Maturity & Settlement	Loan closed, SAG burned	HCS Final Record

---

🧩 6️⃣ Future Extensibility

Silsilat’s modular design allows for:

• Multi-asset expansion: Silver, diamonds, or other tangible assets.

• AI model marketplace: Third-party evaluators can publish models verified by Phoenix and registered on-chain.

• Cross-border liquidity routing: Integration with stablecoins or regional CBDCs.

• ZKP-based privacy proofs: Enable verifiable yet private on-chain KYC attestations.

• DeFi bridges: Liquidity tokens (LQT) can be listed on DeFi exchanges for global participation.