Tasks are the core components of the Rightbrain platform. They are automated processes that execute predefined instructions every time a large language model receives new input data. Each task includes:

• Instructions: What the model should do.

• Dynamic Input Variables: Placeholders (e.g., {customer_review}) that specify which part of the input to process.

• Structured Outputs: Predefined output formats to ensure consistency across runs.

To begin creating a task, click on the Create Task button in the dashboard.

Define your task

First, you’ll have to draft your task in the user prompt. For the purpose of our demo we’ll focus on a sentiment analysis task for a product review. We will enter the follow task in the user prompt field.

User prompt: ‘Please conduct a comprehensive sentiment analysis on the {customer_review} and describe the product image (if one is included). Verify that the image matches the product described in the review. ‘

Here, {customer_review} is a dynamic variable that the model replaces with actual data at runtime. The instruction also tells the model to analyse any provided image.

Choosing your model

Having drafted your task, you have several models to choose from. You can select from powerful advanced reasoning models for complex tasks through to cost-efficient, speedier models for simpler jobs. For this task, you’ll want to choose one with vision capabilities, like Claude Sonnet 3.5 🖼️.

When you’re ready, hit Proceed to go to the next step.

Define Outputs

Specify the outputs you expect from the task by naming them, choosing their data types, and providing a description. This step ensures that every task run returns data in a consistent format for use in your database or app UI.

Go ahead an enter the following outputs for our test:

• Output name - sentiment, type - string, description - Whether the review is positive, negative or neutral

• Output name - image_description, type - string, description - A concise description of the image

• Output name - image_match, type - boolean, description - Whether the image provided matches the product in the review

• Output name - key_terms, type - list, description - A list of key words used in the review

Finalise with a System Prompt

Give your task a name and add a system prompt to provide additional context or constraints (think of it as setting the model’s role). You can also add a webhook to automatically forward task outputs to a specific destination. Here we’ll use the Rightbrain Assistant to help us draft a suitable system prompt.

Just like that, you’ve created your first task on Rightbrain. Now we’re ready to save our task and test it to see if it works.