Generate the Sample Data

The examples in this guide use one dataset from Smart Factory, a fictional plant with eight machines. Generate the Smart Factory files one time in C3 Intelligence, then continue the walkthroughs in the same conversation, where the agents already have the files. The requests below name the Coding Agent, the agent that writes and runs general code and creates documents. You can name another agent, or no agent at all, and the orchestrator routes the work.

Generating the data takes two requests in one conversation:

1. Ask for the seven CSV data files.
2. Ask for the three PDF documents.

Downloading the files is optional. Keep a copy when you want to use the files in another conversation.

The finished set holds seven CSV files of plant data and three PDF documents of plant rules, each summarized in About the data.

Copy link to this section

Create the seven data files

Generate the files in the conversation you plan to work in. To create the data files:

1. Select New conversation in the sidebar, or open an existing conversation, and wait until the message box is ready.
2. Copy this request into the message box, then select Send message:

Use the Coding Agent to create a sample dataset for a fictional plant called Smart Factory, so I
can practice analyzing factory data, predicting breakdowns, and planning two weeks of maintenance
starting July 1, 2025. Cover January 1 to June 30, 2025, and save seven CSV files in one folder
named smart-factory-data:

- machines.csv: the eight machines (three CNC machines, three lathes, and two hydraulic presses),
  with an ID such as M-01, a name, the type, the install year, the floor, and how critical each is.
- technicians.csv: ten maintenance technicians, each with a skill (CNC, general, hydraulic, or
  electrical), a day, night, or weekend shift, and an hourly rate.
- sensor_readings.csv: one reading per machine per hour for temperature, vibration, pressure,
  spindle speed, power, and oil level.
- fault_events.csv: about twenty breakdowns, each with the machine, the fault type, the severity,
  the start and end time, and a short description.
- maintenance_log.csv: about a hundred maintenance jobs, planned, corrective, and emergency, each
  with the technician, the times, the cost, the downtime hours, and, for an emergency repair, the
  breakdown it fixed.
- production_orders.csv: one order per machine per working day, Monday to Saturday, with planned
  and actual times, a status of completed or delayed, the delay in hours, and the product.
- planning_inputs.csv: for each machine, the skill a maintenance visit needs, how many hours the
  visit takes, the cost of each hour of downtime on weekdays and at weekends, the chance of a
  breakdown on any day, and the cost of one breakdown.

Make the data realistic and consistent:

- Three machines break down often: M-02 (a CNC machine), M-05 (a lathe, with bearing problems),
  and M-07 (a hydraulic press). The others break down rarely or never.
- In the days before each breakdown, that machine's readings drift away from normal, such as a
  slow rise in temperature or vibration.
- Emergency repairs cost several times more than planned maintenance, and breakdowns delay some
  production orders.
- An hour of downtime costs less at weekends than on weekdays.
- Leave about one reading in two hundred blank, scattered across machines and dates the way real
  sensor gaps occur. Keep the readings complete during and around breakdowns, so the gaps give no
  hint of when a breakdown happened.
- Use the same machine, technician, and breakdown IDs in every file that refers to them.

Show me your plan before you start. When you finish, list each file with its number of rows and
its date range.

The request has the four parts of a good prompt. It states the outcome, practice data for analysis, prediction, and planning, and sets the scope: seven named files over six months. The request describes each file in plain words and names the patterns the walkthroughs need, but leaves column names and numbers to the agent. A checkpoint asks to see the plan first, and the last sentence says what to return: each file with its row count and date range.

The orchestrator, shown as C3 Intelligence, plans each request and hands the work to the agent the request names, here the Coding Agent. It replies with a plan. Select the option that approves the plan. To change something first, select the option that lets you specify a change, then type it.

The Coding Agent then writes and runs the code that builds the files, which can take several minutes. When it finishes, the reply lists the seven files with their row counts and date ranges. Check that the dates run from January 1 to June 30, 2025, and that the sensor file has about 34,750 rows, one per machine per hour.

Copy link to this section

Create the three documents

The documents give the RAG Agent, the agent that answers questions from your documents, a set of plant rules to search. Write the documents in the same conversation, so they reuse the machine IDs, skills, and costs from the data files. Copy this request into the message box, then select Send message:

Using the Smart Factory data you just created, write three short PDF documents for the same plant,
each two to five pages, and save them in the smart-factory-data folder:

- Equipment-Operation-and-Maintenance-Manual.pdf: the machine list, the normal and warning ranges
  for each sensor reading by machine type, the planned maintenance tasks for each type with their
  length and required skill, and troubleshooting steps and repair times for each fault type in the
  data.
- Preventive-Maintenance-Protocol.pdf: how to plan maintenance in two-week periods with one visit
  per machine, which machines count as high risk, which technician skills and shifts cover the
  work, and the downtime costs.
- Maintenance-Safety-Procedure.pdf: protective equipment, locking out a machine before work, hot
  parts, hydraulic presses, rotating parts, and returning a machine to service.

Keep the machine IDs, skills, visit lengths, and costs the same as in the CSV files. Include one
disagreement for practice: the manual says to let a lathe spindle housing cool for at least 30
minutes before opening it, and the safety procedure says at least 60 minutes and below 40 °C. Show
me an outline of each document first. When you finish, list each document with its page count.

Approve the outline the orchestrator proposes, and wait for the reply that lists the three documents. The request ties the documents to the data and plants the 30 versus 60 minute cool-down conflict on purpose. The Build a RAG Knowledge Base walkthrough asks the RAG Agent to find conflicts like that one.

Copy link to this section

Continue in the same conversation

The ten files are in the smart-factory-data folder of the Coding Agent's workspace, its own storage area in the conversation. The orchestrator moves files between agents, so the next walkthrough can start right here. In Get Started with C3 Intelligence and Analyze Data, send the first request without attaching files, and the agents use the files in this folder.

Your generated data has its own values, so compare your results with the patterns each walkthrough describes rather than its exact figures.

Copy link to this section

Keep a copy of the files (optional)

A copy on your computer lets you attach the files in another conversation. To download the files:

1. In the right panel, select Files.
2. Select Select workspace agent, the icon beside the Files title, then Coding Agent.
3. Open the smart-factory-data folder. If the folder is missing, select Refresh files first.
4. Select Download beside each of the ten files.

To find the folder when it is in another agent's workspace, ask the orchestrator where it saved the files.

Copy link to this section

About the data

The ten files describe the same plant from January 1 to June 30, 2025. The CSV files share machine, technician, and breakdown IDs, so an agent can join them:

Copy link to this section

See also

• Get Started with C3 Intelligence
• Conversation Features