AI Sample Dataset Generator

Overview

Prototype teams often need tabular data before a source system or approved dataset is available. Handwritten rows drift in shape and miss fields that a test or interface needs. This workflow gives you a synthetic dataset in JSONL and CSV, plus a data dictionary for prototypes and tests. Seed Topic defines the subject and constraints, Generate Dataset creates a coherent schema and sample rows, and Finalize Dataset writes the artifacts. Keep the graph with your other visual workflows when you need to regenerate the fixture from a reviewed brief. It creates illustrative records, not company enrichment or facts about real entities.

Pipeline

1. Seed Topic (seed-topic, script) sets the topic, requested row count, and optional schema constraints.
2. Generate Dataset (generate, agent) turns the brief into typed columns, synthetic rows, and a data dictionary.
3. Finalize Dataset (finalize, script) writes matching dataset files and a Markdown data dictionary.

Inputs

• Dataset brief. The topic and intended use for the synthetic records. Default: use the example brief in Seed Topic, then replace it before a production test.
• Row count. The number of sample records requested from the agent. Default: keep the count configured in Seed Topic for a compact fixture.
• Schema constraints. Column names, types, or a data dictionary seed the output must follow. Default: leave this blank so the agent designs the schema from the brief.

Outputs

• JSONL dataset. One synthetic record per line for downstream tests and imports.
• CSV dataset. The same records and columns in spreadsheet-ready form.
• Markdown data dictionary. The schema and each field definition in a reviewable reference.

Integrations

• AI

Sample output

Illustrative output from finalize. These records show the artifact shape and do not describe real equipment, companies, or events.

dataset.jsonl

dataset.csv

asset_id,asset_type,maintenance_state,inspection_window
SYN-001,air-quality-sensor,scheduled,2026-Q4
SYN-002,temperature-sensor,review,2027-Q1

data-dictionary.md

| Column | Type | Meaning |
| --- | --- | --- |
| asset_id | string | Synthetic record identifier |
| asset_type | string | Illustrative equipment class |
| maintenance_state | enum | Scheduled maintenance state |
| inspection_window | string | Illustrative review period |

FAQ

Can this workflow model any tabular topic?

Use subjects that fit rows and columns, such as product catalogs, sensor readings, or survey responses. A specific brief produces a more useful schema than a broad topic name.

Can the agent generate large datasets?

Set the requested count in Seed Topic. Keep each request small enough for one agent response, then split larger fixtures into reviewed batches.

Is the generated data safe to use in production?

The records are synthetic and intended for prototypes and tests. Do not treat them as observed facts or use them as a source for production decisions.

Can I control the schema of the output?

Yes. Add column names, types, or a data dictionary seed in Seed Topic. Leave those constraints blank when you want the agent to propose the schema from the dataset brief.

Related workflows

• Company Enrichment from Website instead when your inputs are real company homepages and you need source-based company attributes.
• Scrape and Summarize Webpages instead when you need concise reading output from real pages rather than synthetic rows.
• Daily Tech Digest instead when you need a recurring briefing from public feeds rather than generated test data.