ResearchAI 1.2.1

💬 ResearchAI is an Azure OpenAI & Computer Vision Client (.NET)

This project provides a simple yet extensible .NET wrapper for Azure OpenAI's Chat Completion API, including:

• Per-session chat memory (thread-safe)
• Configurable ChatCompletionOptions (like temperature, penalties, token limits)
• Injection of system context
• Clean abstraction for chatbot/message processing
• Analyze images using Azure Vision API

🚀 Features

• ✅ Session-based memory (thread-safe with ConcurrentDictionary)
• ✅ Configurable model behavior via ChatCompletionSettings
• ✅ Custom system context per session
• ✅ Automatically limits and validates Azure AI parameters
• ✅ Analyze images using Azure Vision API

🔧 Configuration

Here’s how you can configure the client in your application:

{
  "AzureOpenAI": {
    "Endpoint": "https://<your-openai-resource>.openai.azure.com/",
    "Key": "<your-api-key>",
    "DeploymentName": "gpt-35-turbo",
    "SystemContext": "You are a creative assistant for generating photo captions.",
    "ChatCompletionSettings": {
      "Temperature": 0.8,
      "MaxOutputTokenCount": 800,
      "FrequencyPenalty": 0.2,
      "PresencePenalty": 0.3
    }
  },
  "ImageVisionAI": {
    "Endpoint": "https://<your-visionai-resource>.cognitiveservices.azure.com/",
    "Key": "<your-api-key>"
  }
}

🧪 Sample Usage

var options = new AzureOpenAIOptions
{
    Endpoint = "<your-endpoint>",
    Key = "<your-key>",
    DeploymentName = "gpt-35-turbo",
    SystemContext = "You are a helpful assistant.",
    ChatCompletionSettings = new ChatCompletionSettings
    {
        Temperature = 0.7f,
        MaxOutputTokenCount = 800,
        FrequencyPenalty = 0.0f,
        PresencePenalty = 0.0f
    }
};

var openAI = new AzureOpenAI(options);

var response = await openAI.SendMessageAsync("user-session-id", "Describe a sunset wedding on the beach.");
Console.WriteLine(response.Content[0].Text);

🧠 How It Works

• First message initializes a session with the system prompt.
• Each user message is added to the session history.
• The history is passed to Azure OpenAI to maintain conversation context.
• The assistant response is added back into the session.

📸 Analyze Image

var imageClientModel = new ImageClientModel
{
    Endpoint = "<your-computer-vision-endpoint>",
    Key = "<your-computer-vision-key>"
};

var client = new ImageClient(imageClientModel);
var result = client.AnalyzeImage("https://example.com/image.jpg");

With Azure OpenAI (summary, caption, tags, etc.)

string output = await client.AnalyzeImageWithChatAsync("https://example.com/image.jpg", "your-session-id");
Console.WriteLine(output);

🧾 Sample Output Format from Image Client

{
  "title": "City Walk Vibes",
  "caption": "A group of friends enjoying a sunny day in the city, dressed casually with smiles all around.",
  "instagram_tags": ["#citylife", "#friends", "#urbanstyle", "#weekendvibes"],
  "facebook_tags": ["#CityWalk", "#HappyPeople", "#CasualLook"],
  "objects": ["person", "building", "road", "backpack"],
  "people": [
    {
      "gender": "male",
      "clothing": "blue jeans and white t-shirt",
      "description": "Man walking with a smile carrying a backpack",
      "activity": "walking in the city"
    }
  ]
}

📌 Notes

• Temperature and Penalty values are clamped to valid ranges (0–2).
• MaxOutputTokenCount is limited between 512 and 800 to fit model constraints.
• You must provide a valid sessionId for session tracking to work.

📄 License

Apache 2.0 License