.NET AI Structured Output Generator

Effortlessly query AI models (starting with Google's Gemini) and receive strongly-typed, structured C# objects as output. Instead of parsing messy JSON or string responses, define your desired C# class, and let the AI fill it in for you!

Features

• Strongly-Typed Outputs: Get C# objects directly from AI responses.
• Attribute-Driven Schema: Use [Description] attributes on your class properties to guide the AI in populating the data.
• Data Type Support: Handles common types like string, decimal, DateTime, enum, double, bool, etc.
• Constraints: Use attributes like [Range] to provide further guidance to the AI.
• Easy Integration: Simple setup with .NET's dependency injection.
• Provider-Based: Currently supports Google Gemini, with a design that can be extended to other AI providers.

Prerequisites

• .NET 6.0 or later.
• A Google Gemini API Key. You can obtain one from Google AI Studio.

Installation

Install the package via NuGet Package Manager:

Install-Package YourPackageName

Or via the .NET CLI:

dotnet add package YourPackageName

(Remember to replace YourPackageName with your actual package name once published!)

Getting Started

1. Configure Services

In your Program.cs or startup configuration, add the Gemini structured output generator:

using AI.StructuredOutput; // Assuming this is your root namespace using Microsoft.Extensions.DependencyInjection; using Microsoft.Extensions.Hosting; public class Program { public static async Task Main(string[] args) { using IHost host = Host.CreateDefaultBuilder(args) .ConfigureServices(services => { // Replace "YOUR_GEMINI_API_KEY" with your actual API key services.AddGeminiStructuredOutputGenerator("YOUR_GEMINI_API_KEY"); }) .Build(); await host.StartAsync(); // ... your application logic } }

Important: Securely manage your API key. Consider using user secrets, environment variables, or a configuration provider for production applications.

2. Define Your Output Structure

Create a C# class that represents the structure you want the AI to populate. Use [Description] attributes to guide the AI on what each property means.

using System.ComponentModel; using System.ComponentModel.DataAnnotations; // For [Range] public class WheatherInfo // Renamed from WheatherInfo for correct spelling { [Description("The city and country, e.g., 'London, UK' or 'Paris, France'")] public string Location { get; set; } [Description("The current temperature in Fahrenheit")] public decimal Fahrenheit { get; set; } [Description("The current temperature in Celcius")] public decimal Celcius { get; set; } [Description("A very short, friendly greeting or a brief, light-hearted comment about the weather. For example: 'Enjoy the sunshine!' or 'Stay dry!' Do not make a joke, just say something like ´good morning´")] // Clarified instruction for AI public string Joke { get; set; } // Consider renaming to `FriendlyMessage` or similar if "Joke" is misleading [Description("The local date and time at the specified location when these weather conditions were recorded")] public DateTime TimeStamp { get; set; } [Description("The current wind conditions")] public WindCondition WindCondition { get; set; } [Description("Assign a random decimal number between 0.0 and 1.0 (inclusive)")] [Range(0.0, 1.0)] public double Test { get; set; } // Perhaps rename to `RandomValue` for clarity } public enum WindCondition { [Description("Calm, almost no wind")] NoWind = 0, [Description("A gentle breeze is present")] LightWind = 1, [Description("Noticeably windy, potentially strong gusts")] StrongWind = 2, }

Tip: The more descriptive your [Description] attributes are, the better the AI will understand what data to provide. You can also guide the format of the data.

3. Query the AI

Retrieve the IAiStructuredOutputGenerator service and use the AskAsync<T>() method:

// (Continuing from Program.cs Main method or another part of your application) var service = host.Services.GetRequiredService<IAiStructuredOutputGenerator>(); string? city = null; while (string.IsNullOrWhiteSpace(city)) { Console.Write("Type a city to query the current weather: "); city = Console.ReadLine(); } Console.WriteLine($"Fetching weather for {city}..."); var weatherInfo = await service.AskAsync<WheatherInfo>($"What´s the weather like in {city}?"); if (weatherInfo == null) { Console.WriteLine("Oops, something went wrong. The AI couldn't provide structured data."); } else { Console.WriteLine($"--- Weather in {weatherInfo.Location} ({weatherInfo.TimeStamp}) ---"); Console.WriteLine($"Temperature: {weatherInfo.Celcius}°C / {weatherInfo.Fahrenheit}°F"); Console.WriteLine($"Wind: {weatherInfo.WindCondition}"); Console.WriteLine($"AI's friendly note: {weatherInfo.Joke}"); Console.WriteLine($"Random Test Number: {weatherInfo.Test}"); } Console.Write("Press any key to close..."); Console.ReadKey();

Example Output

Type a city to query the current weather: London Fetching weather for London... --- Weather in London, UK (2023-10-27 10:30:00) --- Temperature: 12°C / 53.6°F Wind: LightWind AI's friendly note: Good morning! Random Test Number: 0.734 Press any key to close...

(Note: Actual output will vary based on the AI's response.)

How it Works (Briefly)

The SDK takes your C# class definition and uses the property names, types, and [Description] attributes to construct a schema or a set of instructions for the AI. It then sends your prompt along with these instructions to the configured AI model (e.g., Gemini). The AI attempts to generate a response that conforms to the requested structure, which the SDK then deserializes into an instance of your C# class.

Supported AI Providers

• Google Gemini: Configured using services.AddGeminiStructuredOutputGenerator("YOUR_API_KEY").

(Future providers can be added by implementing IAiStructuredOutputGenerator and providing a corresponding extension method.)

License

Free for all