Skip to content

README.md

Latest commit

 

History

History
94 lines (69 loc) · 4.19 KB

README.md

File metadata and controls

94 lines (69 loc) · 4.19 KB

SmartLend 🏦

AI-Powered Financial Loan Decision Engine

Enterprise-grade Loan Processing System built with Clean Architecture, Domain-Driven Design (DDD), and .NET 10.

🚀 Overview

SmartLend is a compliance-ready financial engine designed to automate loan risk assessment. It moves beyond simple CRUD to demonstrate how Deterministic Logic and Generative AI can work together in a secure banking environment.

The system enforces strict financial rules using C# (safe), while leveraging Google Gemini 2.5 Flash to generate empathetic, human-readable explanations for customers (smart).

🏗️ Architecture & Design Decisions

The solution follows strict Clean Architecture principles to ensure that Business Logic is isolated from Technical Details.

The 4 Layers (The Onion)

  1. 🟡 Domain (The Core): Contains the Enterprise Business Rules.
    • Rich Domain Model: Logic is encapsulated in LoanApplication.Evaluate(), ensuring a loan can never exist in an invalid state.
    • Pure: Zero dependencies on external libraries.
  2. 🔴 Application (The Orchestrator): Contains the Use Cases.
    • Orchestrates the flow: DTO -> Domain Logic -> Database -> AI Advisor.
  3. 🔵 Infrastructure (The Plugins): Contains the external tools.
    • Persistence Strategy: Implements the Repository Pattern with Dual Support (In-Memory & SQL Server).
    • AI Service: Google Gemini 2.5 Flash implementation via HttpClient.
  4. 🟢 API (The Entry Point): The REST API/Swagger UI.
    • Wires everything together using Dependency Injection.

🌟 Key Features

🤖 1. AI Financial Advisor (Google Gemini 2.5)

I integrated Google Gemini 2.5 Flash into the Infrastructure layer to act as a "Financial Advisor."

  • Workflow: When a loan is rejected by the C# math engine, the AI analyzes the user's data (Salary, Credit Score) and generates a polite, personalized financial tip.
  • Safety: The AI never makes the approval decision. It only explains it. This prevents "AI Hallucinations" from approving risky loans.

💾 2. Dual Persistence Strategy (SQL & In-Memory)

To demonstrate architectural flexibility, I implemented two database providers in the Infrastructure layer:

  • InMemoryLoanRepository: Optimized for rapid development, unit testing, and demos (Currently Active).
  • SqlServerLoanRepository: Production-ready implementation using Entity Framework Core.
  • Note: The application is currently configured to run in In-Memory Mode for ease of review (no local SQL setup required). To switch to SQL, simply uncomment the line in Program.cs.

📜 3. Compliance Audit Log

Every decision is immutable. The system records a timestamped (UTC) Audit Log for every action.

  • Example Log: [AUDIT] Action: Rejected | Time: 12/05/2025 | AI Advice: "Debt-to-Income ratio too high..."

🛠️ Tech Stack

  • Framework: .NET 10 (Preview)
  • Architecture: Clean Architecture / Onion Architecture
  • AI Model: Google Gemini 2.5 Flash
  • Database: In-Memory (Active) / SQL Server (Implemented)
  • Testing: xUnit
  • Documentation: Swagger / OpenAPI

🏃‍♂️ How to Run

Prerequisites

  • .NET 10.0 SDK installed.
  • (Optional) A Google Gemini API Key. Note: If no key is provided, the system gracefully degrades to "Offline Mode" and still processes loans.

Steps

  1. Clone the repository:

    git clone [https://github.com/TaGoat/SmartLend.git](https://github.com/TaGoat/SmartLend.git)
cd SmartLend

  2. Run the API:

    dotnet run --project SmartLend.Api

  3. Test with Swagger: Open your browser to the URL shown in the terminal (e.g., http://localhost:5xxx/swagger).

    • POST /api/loans: Submit a loan request.
    • Check Console: Watch the VS Code terminal to see the AI Advice and Audit Logs printed in real-time.

🧪 Testing Strategy

Core business logic is covered by Unit Tests in the SmartLend.Domain.Tests project.

  • Run Tests: dotnet test
  • Coverage: Verifies that the "Debt-to-Income" and "Credit Score" rules trigger the correct Rejection status.

Built by [Tareq AL-Suraihi]