ETLBox ETLBox ETLBox Buy Try Buy

Chat Transformation

This article contains example code that shows the usage of the ChatTransformation component.

ChatTransformation sends each row as JSON to an IChatClient and writes the returned object downstream. A typed output uses a JSON schema derived from the output type. ExpandoObject output uses JSON mode.

Classify a typed object

This example classifies product reviews. Id, Text, and CustomerCode stay on the row. The model adds Sentiment and Score.

public enum Sentiment
{
    Negative,
    Positive
}

public class Review
{
    public int Id { get; set; }
    public string Text { get; set; }
    public string CustomerCode { get; set; }
}

public class ReviewLabel
{
    public int Id { get; set; }
    public string Text { get; set; }
    public string CustomerCode { get; set; }
    public Sentiment Sentiment { get; set; }
    public int Score { get; set; }
}

var source = new MemorySource<Review>();
source.DataAsList.Add(new Review { Id = 1, Text = "I love this product.", CustomerCode = "C-100" });
source.DataAsList.Add(new Review { Id = 2, Text = "This is broken and useless.", CustomerCode = "C-200" });
source.DataAsList.Add(new Review { Id = 3, Text = "Great quality and fast delivery.", CustomerCode = "C-300" });
source.DataAsList.Add(new Review { Id = 4, Text = "Poor quality, I want my money back.", CustomerCode = "C-400" });

var client = new OpenAIClient(openAIApiKey)
                .GetChatClient(openAIModel)
                .AsIChatClient();

var chat = new ChatTransformation<Review, ReviewLabel>(client) {
    SystemPrompt = "You classify product reviews.",
    UserPrompt = "Keep Id, Text, and CustomerCode unchanged. Set Sentiment to Positive or Negative and add Score as an integer from 0 to 100."
};
var dest = new MemoryDestination<ReviewLabel>();

source.LinkTo<ReviewLabel>(chat).LinkTo(dest);
Network.Execute(source);

foreach (var row in dest.Data)
    Console.WriteLine($"Id:{row.Id} CustomerCode:{row.CustomerCode} Sentiment:{row.Sentiment} Score:{row.Score}");

//Outputs (might vary!)
//Id:1 CustomerCode:C-100 Sentiment:Positive Score:95
//Id:2 CustomerCode:C-200 Sentiment:Negative Score:10
//Id:3 CustomerCode:C-300 Sentiment:Positive Score:95
//Id:4 CustomerCode:C-400 Sentiment:Negative Score:20

Classify a dynamic object

The same classification works with ExpandoObject. The non-generic ChatTransformation reads and writes dynamic rows.

var source = new MemorySource();
string[] texts = {
    "I love this product.",
    "This is broken and useless.",
    "Great quality and fast delivery.",
    "Poor quality, I want my money back."
};
for (int i = 0; i < texts.Length; i++) {
    dynamic review = new ExpandoObject();
    review.Id = i + 1;
    review.Text = texts[i];
    source.DataAsList.Add(review);
}

var client = new OpenAIClient(openAIApiKey)
                .GetChatClient(openAIModel)
                .AsIChatClient();

var chat = new ChatTransformation(client) {
    SystemPrompt = "You classify product reviews.",
    UserPrompt = "Keep Id and Text unchanged. Add Sentiment set to exactly Positive or Negative and Score as an integer from 0 to 100."
};
var dest = new MemoryDestination();

source.LinkTo(chat).LinkTo(dest);
Network.Execute(source);

foreach (dynamic row in dest.Data)
    Console.WriteLine($"Id:{row.Id} Text:{row.Text} Sentiment:{row.Sentiment} Score:{row.Score}");

//Outputs
//Id:1 Text:I love this product. Sentiment:Positive Score:95
//Id:2 Text:This is broken and useless. Sentiment:Negative Score:5
//Id:3 Text:Great quality and fast delivery. Sentiment:Positive Score:90
//Id:4 Text:Poor quality, I want my money back. Sentiment:Negative Score:10
Edit this page on GitHub
Prev
Chat Batch Transformation
Next
Embedding Batch Transformation