DetaBord Demo Instances

We have several DetaBord demo instances running on our own clusters. Please contact us at to request access.

The demo instances run in a special sandbox mode with limited functionality. This page provides a sample walkthrough on how to best use the DetaBord demo.


As DetaBord is completely end-to-end encrypted between clients and platform (see architecture for details) normally every DetaBord user must install the DetaBord CLI including the web application locally on their machines. For easy access the DetaBord demo instances instead offer a number of ready-to-use CLI instances, the web applications of which can be accessed directly online via any browser.

Use the provided URLs and credentials to log in to the demo CLI instances. You can log in with a “user” or “data owner” role, for data analysis or machine learning development and auditing or approval management respectively.

To start, log in as a “user”.

Demo Data

The Demo instances contain two pre-configured data sets, each in three different privacy levels. for you to use. The DetaBord demo sandbox mode prevents editing, removing, or adding data sets.

The two datasets are:

  • Census Adult dataset
  • TPC-H benchmark dataset
Dataset NamePrivacy LevelUsage
Adult_Census_Income_With_Header_PublicPublic“ML”, “Estimator”, and “Synth” runs
Adult_Census_Income_With_Header_Privacy-1High“ML”, “Estimator”, and “Synth” runs, full run results require approvals
Adult_Census_Income_With_Header_Privacy-2Highest“ML”, “Estimator”, and “Synth” runs, all run results require approvals
TPC-H_Small_PublicPublicSQL Queries
TPC-H_Small_Privacy-1HighSQL Queries, full run results require approvals
TPC-H_Small_Privacy-2HighestSQL Queries, all run results require approvals

Demo Use

There are five use cases of the DetaBord platform that you can explore in the demo:

  • Machine Learning (ML) Runs
  • Estimator Runs
  • SQL Queries
  • Synthetic Data Runs
  • Privacy Checker

The starting point of every run in DetaBord – except for SQL queries – is your local workspace. The demo instance comes with a prepared demo workspace that is not editable but contains examples for all four use cases (the privacy checker use case can be accessed differently). Go to the workspace by clicking on the “Demo User Workspace” link on the top:

Demo Workspace

Machine Learning

The prepared machine learning use cases trains a neural network classification model on the Census adult dataset. Learn more about this example here: ML example

Go to your demo workspace and select the folder named “ML” in the workspaces’ file explorer. Next, click the “Prepare new run with this folder” button to select this folder for your next run. Make sure to deselect the current folder selection first by clicking on the cross on the right-hand side, if this button is not visible.

Workspace ML

Once you clicked the “Prepare new run…” button the “Start Run” component is shown in the lower part of the screen:

Start Run ML

Select an experiment for this run and the dataset you want to use for this model training run. For the demo leave the entry point and epsilon parameters as is. As mentioned above only the Census dataset can be used for this type of run in the DetaBord demo environment. Selecting a different dataset or multiple dataset is possible but will result in runs completing in error states.

Depending on whether you choose dataset “…_Public”, “…_Privacy-1”, or “…_Privacy-2” the run results will immediately be accessible or must be requested for approval from the data owner.

Click “Start Run” to start the training run and automatically navigate to the run’s details page. Go to the “Files” tab to inspect the logs of the run:

Run ML

If you trained on a non-public dataset, you the “Get Approvals” button to request access to all the results. Learn more about approvals here: Approvals


The estimator use case is similar to the machine learning use case above. Only here pre-defined estimator models can be used.

Again, start by navigating to the “Demo User Workspace”. The select the “Estimator” folder and click the button “Prepare new run with this folder”.

Unlike with the general machine learning use case the estimator runs will not use user-defined model code, but rather pre-defined estimator classes that can be configured as described in the MLProject file:

Start Run Estimator

Use the “Entry Point” dropdown to select a suitable estimator class:

Start Run Estimator Entry Point

Click “Start Run” and inspect the run details as described above.

SQL Queries

As with any computation task on the DetaBord platform SQL Queries are managed by DetaBord as normal mlflow runs. Therefore a “Query” template folder exists in the demo workspace. Nevertheless, for use of use, SQL queries are handled differently on the frontend side. The “Query” template folder is created automatically (in non-sandbox DetaBord instances) and can be ignored. To start a SQL Query rather navigate to the “Queries” menu item:

Start Run Query

Select the TPC-H dataset from the dropdown on the right and enter a valid query in the query editor. You can use the pre-defined sample query in this sandbox demo mode by clicking on the “Sandbox mode: Add ready to use SQL statement” button after selecting the empty query editor:

Query 1
Query 2

Click on “Run” and navigate to the run details page by selecting the query run from the “Latest Queries” list below.

Query Results

Synthetic Data

Generating synthetic data can be useful to learn more data sets that are private and therefore unavailable for direct inspection.

The synthethic data generation built into DetaBord consists of two parts: training the synthetizer model and generating the synthesized data. In this sandbox mode only the first step is available (because the creation of new data sets are prevented in the sandbox).

Go to the “Demo User Workspace” and select the “Synth” folder. Prepare the folder for a run and select a dataset for the the synth training model. In the demo instance the Census dataset can be used for synthesizer training:

Synth Run

Privacy Checker

The Privacy Checker is a special run that can be performed by the data owner to learn more about the privacy properties of a trained machine learning model.

Log in as a “owner” user and navigate to the “Runs” page. Use the “Settings” dropdown on the right to select the “” filter. This will give you all the runs of this user.

User Runs

In the list select a run of type “Model Train” and go to the “Files” tab. Below the “Assets” in the file explorer you will find a button called “Run Privacy Checker”:

Run Privacy Checker

Press this button and select the “md_classification” template. Further configure the privacy checker as described here: Privacy Checker.

When you started the privacy checker run, navigate back to the “Runs” list and select the privacy checker run to inspect the results in the “Files” tab:

Privacy Checker Results

Join Waitlist

Join the AI for Life Sciences Challenge

Let's talk