.svg)
.svg)
Today, we’re announcing our JSON Task in the Arcus Task Playground. The JSON Task uses Arcus’ agent-native platform to produce structured data from complex enterprise data, automating tedious, manual data-processing.
The JSON Task is designed to be plugged into repeated workflows. Once a JSON Task has been configured in the Arcus Task Playground, it can be run over different data sources to perform the same task over and over again. Whether it's programmatically integrated into internal business logic, or used directly in the Arcus Task Playground, the JSON task allows customers to seamlessly leverage the power of the Arcus Platform in their workflows.
The biggest challenge with data-intensive business workflows is the sheer quantity and complexity of data that needs to be processed accurately and efficiently. In order to ensure the highest accuracy for the fields in the JSON Task, the JSON Task leverages Arcus’ multimodal data understanding capabilities and planet-scale retrieval systems to generate each field that the user specifies. This means that the fields can be extracted and generated across massive data volumes containing a variety of data modalities, such as complex tables, figures, charts, images, etc. Finally, the fields are outputted in an aggregated, structured JSON format according to an input schema defined by the user, allowing for complete control and customizability over the response format.
Each field of the JSON Task is not only powered by Arcus’ data understanding capabilities but also leverages Arcus’ AI-agents and tool-use abilities. This ensures that the complexity of each field is handled with the highest accuracy possible. For example, some fields may require mathematical computation, while other fields may need to reconcile internal business data with up-to-date information from the internet. Through leveraging Arcus’ agents to extract and generate each field, the JSON Task greatly reduces the amount of manual effort required for these processing tasks. This makes the JSON task an ideal solution for high volume, high accuracy, repeated workflows.
With our Show-Your-Work feature, users get to stay in the drivers seat while using the JSON Task. This feature allows users to trace exactly how Arcus reached its answers for each field, what information from the data source was used to inform the answer, and what external information was pulled in to inform the answer. This not only gives users unparalleled visibility into the generation process at a per-field level, but also allows users to have confidence in the answers and empowers them to perform any sanity checks as necessary.
Let’s go through an example of using the JSON Task. Imagine conducting a sweeping Compliance and Regulatory audit across ESG disclosures from all major Fortune 500 tech companies. Part of this audit process might involve extracting values like the following:
First in the Task Playground, the JSON Task can be configured to extract these three fields: the company overview, the sustainability strategies and results, and the PUE. Each field in the JSON Task can be configured according to numerous parameters such as a prompt, field description, field examples, model guidance, and other keys related to this field that should be extracted. The JSON Task also supports defining specific response types (e.g. percentages, dates, integers, etc.) for each field to ensure that the response JSON is compatible with any downstream tasks and aligns with custom data models.
By configuring these parameters for each field, the JSON task can be tuned to your exact domain and exact business context. For example, if only European data centers are relevant for this audit, all that is needed is a sentence in the task configuration asking the model to limit the PUE field to European data centers only.
Our JSON Task also supports hierarchical, multistep extraction through our Entity field type. A common step in a data extraction workflow follows the following format: first, find all “X” over a given set of data, then for each “X”, extract another set of values that correspond to that “X”. Our Entity fields are designed to support this kind of step within a workflow. In our ESG example, the sustainability strategy field should be configured to be an Entity, meaning that first Arcus will parse all of the data to find all of the sustainability strategies for the company, then for each sustainability strategy, it will respond with the relevant details for that strategy and the results of that strategy.
Now that the ESG JSON Task has been configured, it can be repeatedly run over all of the ESG reports to extract these relevant fields for each of the companies that need to be audited. What would’ve been a tedious, manual, labor-intensive process of parsing and combing through each company’s ESG disclosures can now be easily automated with the click of a button using our JSON Task.
The JSON Task is already being used by our customers across numerous different industries including financial services, logistics, and insurance. To learn more about what the JSON Task can do for you, reach out to request a demo.
We’re building a reality where you can focus on the work that matters most, and AI takes care of the rest. Reach out if you’re interested in joining our team!
.svg)
