Secure AI Vision Integration (C#)

This repository demonstrates secure multimodal AI vision integration from C#. It is a small portfolio slice for teams that want to add document/image interpretation to an agentic workflow without hardcoding secrets, binding business logic to one provider, or making live model calls impossible to test.

The sample is designed to plug into an autonomous coding harness such as loop-design-build, a standard CI/CD pipeline, or a conventional .NET application boundary.

Portfolio Narrative

The business value is not the vision call by itself. The project shows the surrounding engineering discipline needed for corporate use: environment-based secrets, mockable AI boundaries, deterministic test paths, and a clean interface that lets orchestration code call vision capabilities without owning credentials or provider-specific request details.

It illustrates a useful enterprise pattern: let the model interpret visual input, but keep configuration, invocation policy, error handling, and downstream decisions in ordinary C# code.

🎯 Key Objectives

1. Secure by Default: Avoids hardcoding any API strings or secrets. Credentials are systematically extracted from environment variables (mapping cleanly to CI/CD Vaults/Secrets).
2. Dependency Injection Ready: The IVisionAgent interface and HttpClient usage are structured to directly plug into standard .NET Core DI containers.
3. Agentic Harness Friendly: The code architecture is decoupled, allowing an orchestration agent to abstract the AI interactions and run unit tests deterministically.

⚙️ Configuration

The application uses the standard .NET configuration provider, which loads settings from multiple sources in the following priority (last one wins):

1. appsettings.json: Base settings for the endpoint and model.
2. User Secrets: Best for local development API keys.
3. Environment Variables: Best for CI/CD and production.

Available Settings

Key	Description	Default / Example
AI_VISION_API_KEY	Your AI provider API Key.	(Required for real inference)
AI_VISION_ENDPOINT	The REST API endpoint for the vision model.	https://api.example.com/v1/...
AI_VISION_MODEL	The specific model string to use.	vision-model-v1
AI_VISION_USE_MOCK	Forces deterministic mock responses for demos/tests.	true / false

---

🚀 Getting Started

Prerequisites

• .NET 8.0 SDK

Step 1: Configure your API Key (Local Development)

The easiest way to set your API key locally without touching files is using User Secrets:

cd CSharpVisionAI
dotnet user-secrets init
dotnet user-secrets set "AI_VISION_API_KEY" "your-actual-api-key"

Step 2: Run the Web Demo

# Start the web server
dotnet run

Note

If AI_VISION_API_KEY is omitted, the demo will automatically inject a dummy key and use a mock response to ensure the UI remains testable without a real provider.

Once started, open your browser and navigate to http://localhost:5000. You can drag-and-drop an image and submit a prompt to see the secure vision analysis in action.

Running Tests

The solution includes a dedicated test harness to verify the API boundary and project logic:

cd CSharpVisionAI.Tests
dotnet run

🏗️ Architecture

public interface IVisionAgent
{
    Task<string> AnalyzeImageAsync(string imagePath, string prompt);
}

The application leverages standard System.Net.Http along with System.Text.Json to communicate with standard AI Model REST APIs, ensuring zero heavy vendor-locked SDKs unless absolutely necessary for your business logic domain.

Future Production Ideas

See docs/state/backlog.md for icebox notes on evolving the synchronous demo into a queued, observable AI job workflow with status tracking.