Skip to main content
A data job turns your documents into a training dataset you can use to fine-tune a model. You create a job in two stages. First you name the job and describe the data you want. Then you add a training prompt and files, and start the job when you’re ready. You can leave after the first stage and return later to finish setup.
This page covers the app. To create data jobs in code, see Create instruction fine-tuning data in the SDK guide.

Create a job

1
Navigate to Data Engine > AI-Ready Data.
2
Click Create Job.
3
In Name, enter a name for the job. Names can be up to 100 characters.
4
In Data Job Instructions, describe how to generate question-and-answer pairs from your data. Include the use case, the types of questions, the answer style, and any constraints. Instructions can be up to 10,000 characters. For example: “Generate a training dataset for a customer support assistant on product documentation. Cover common user questions about setup, billing, and troubleshooting.”The Create AI-Ready Data Job dialog with the Name and Data Job Instructions fields and a character count under each.The Create AI-Ready Data Job dialog with the Name and Data Job Instructions fields and a character count under each.
5
Click Create Job.
6
Click Go to Job to open the job’s summary page and finish setup.
You can edit the name and instructions on the summary page until you start the job. The AI-Ready Data Job Summary page showing Job Details, the Training Prompt Setup options, and the Start Job button. The AI-Ready Data Job Summary page showing Job Details, the Training Prompt Setup options, and the Start Job button.

Add a training prompt

The training prompt guides how the model responds in the generated question-and-answer pairs. A training prompt is optional. If you don’t provide one, SeekrFlow generates one from your data job instructions when the job starts. On the job’s summary page, go to Training Prompt Setup and choose Generate Prompt or Write Prompt.

Generate a prompt

1
Click Generate Prompt.
2
In the Generate a Training Prompt dialog, describe how you want the model to behave. Instructions can be up to 1,000 characters.
3
Click Generate.
SeekrFlow writes a prompt from your instructions. The Generate a Training Prompt dialog with a Training Prompt Instructions field, a character count, and a Generate button. The Generate a Training Prompt dialog with a Training Prompt Instructions field, a character count, and a Generate button.

Write a prompt

1
Click Write Prompt.
2
In the Enter a Training Prompt dialog, write a prompt or paste one. Describe the role, tone, response style, and formatting. Prompts can be up to 1,000 characters.
3
Click Continue.
The Enter a Training Prompt dialog with a Training Prompt field, a character count, and a Continue button. The Enter a Training Prompt dialog with a Training Prompt field, a character count, and a Continue button.

Change a prompt

Before you start the job, click the edit icon next to Training Prompt.

Add files

Before you add files, check them against the formats and structure in Prepare files for upload.
1
On the job’s summary page, go to Data Sources and click Add Files.
2
Choose where the files come from:
  • Select Files – Select files you already uploaded to storage.
  • Upload Files – Drag and drop files, or click Browse. You can upload up to 20 files at a time, each up to 150 MB. Accepted formats: PDF, Word, PowerPoint, Markdown, and text.
3
Click Add Files.
The Data Sources section of a data job with an Add Files button and a table of files, each with a status and size. The Data Sources section of a data job with an Add Files button and a table of files, each with a status and size.

Review converted files

SeekrFlow converts files to Markdown automatically when needed. The Data Sources table shows each file’s status and size. To review a converted file, download it. To change it, edit it, upload the updated version, and remove the original.

Start the job

1
Click Start Job. The button becomes available when the job’s status is Ready to Start.
2
In the confirmation dialog, click Start Job.
The status changes to Queued, and SeekrFlow locks the job’s instructions, prompt, and data sources. A job can take a few hours, and SeekrFlow emails you when it finishes. To follow the job and review its output, see Manage data jobs. The summary page of a queued data job, with the Cancel Job button, a Queued status, and an Event Timeline that includes Alignment Job Created. The summary page of a queued data job, with the Cancel Job button, a Queued status, and an Event Timeline that includes Alignment Job Created.
Last modified on November 19, 2026