How to Build AI Conversational Assistants With Zelros and Dataiku Platforms

At Zelros, we are on a mission : to become the conversational layer for critical enterprise business processes.

Natural language interfaces put the power of machine learning systems in the hands of their human users: that’s what’s needed for AI’s wide adoption, and that’s we do.

Our award winning platform helps organizations reinvent how they work, by creating intelligent virtual assistants. Everywhere our solution is deployed, employees can leverage AI to save time for decision making — without compromising on ethics.

We want our product to integrate smoothly with our customers IT systems. That’s why we work hard to add connectors with more and more professional data sources (SAP, Salesforce, Database Systems, Excel, …), third-party data intelligence tools (Tableau, Splunk, …), and enterprise instant messengers (Skype for Business, Microsoft Teams, Slack, Google Hangout, …)

In this post we will demonstrate how Zelros platform can work with Dataiku DSS, the famous and stunning collaborative data science platform, to build state-of-the art predictive assistants.


You are the manager of the largest branch of a big car-rental company. One of your main challenge is to optimize your fleet of 200+ vehicles, to avoid down times. You must replace cars that are likely to have technical problems. But which ones ?

Your company’s data science team uses Dataiku DSS to build predictive models. Thanks to historical data, they can compute a daily risk score for each car of your park.

At the end of the day, predictive data is available through excel-like tables:

Your Wish

Several times a day, during your car review tour, or when proposing a car to an important customer, you ask yourself :

“is this car at risk ? Should I select another one ?”

You are on the field, and the answer is probably somewhere in an excel spreadsheet, buried in you emails. But it would be more convenient if you could ask this question in real-time to your smartphone 

A predictive assistant in your pocket

Here is how to configure Zelros platform to build in a minute a smart assistant powered by Dataiku DSS data.

Step 1: create a new Zelros assistant skeleton

Zelros assistants come out of the box with basic small talk capabilities like hello / goodbye / thank you.

You can configure some other important pieces of the assistant like the help function, the tips, or the failover to a human channel.

However at this step, the assistant is not yet fuelled by relevant data, and remains fairly dumb 😬

Step 2: connect Zelros with Dataiku DSS

Dataiku DSS and Zelros support several data connections types. Good news, both support SQL Database PostgreSQL. Let’s choose this one!

Step 3: configure your first question

The first question you want to serve with the assistant is: “Is the car AA111AA at risk ?”.

It’s pretty easy to support this question. In Zelros administration interface you just have to:

  • add a new intent of type ‘SQL’ (we support other types of advanced intents, learn more about them in future blog post 😉)
  • configure a few utterances like “what is the risk of car @refcar”, “is the car @refcar at risk
  • configure in the field ‘query’ the SQL command to get the answer

Step 4: connect to your preferred Instant Messenger

We support all professional Instant Messengers (e.g. Skype for Business), and adapt gracefully to their specificities.

Here you are: a new Zelros assistant comes auto-magically on your smartphone — nothing to install — with robust Natural Language Understanding, smart entity detection, notifications, next-best-action buttons, and rich formatting.

Congrats, you’re done! Your first assistant is up and running!

And now, enjoy!

You now have a 24/24 7/7 assistant in your pocket, helping you to take the best data-driven decisions for your business.

But you still have important questions, not answered yet, like:

which cars are most at risk ?

No issue — just ask the question! Zelros platform will process unresolved intents, cluster them, and surface them in the AI education interface of the platform. You can now add this new capability by following the previous steps!

What’s next

Our B2E (Business to Employees) platform can be used to optimize many business processes as critical as industrial asset management, finance operations, procurement, customer support, … We have clients in industries like banking, insurance, energy, utilities, … If you want to schedule a demo in your context, drop us an email!

And by the way… we’re hiring!