Dataiku recently introduced their Cobuild feature, which brings AI chat right into the platform. Dataiku describes it as, “The natural-language way to work across Dataiku. Get trusted answers from your data, build governed AI, and oversee every agent you run. One interface, across your whole platform.” At the Succeed conference in NYC in September, it was touted as forever changing the way users interact with Dataiku, even claiming that users will always want to build conversationally going forward. I’ve had a few weeks to experiment with Cobuild and I agree with Dataiku’s predictions mostly, but not entirely.
I found amazing examples of where Cobuild saved me time and did things better than I could do them myself. Here are my top 5:
- Creating a web app. In the past, this was laborious work. Writing code myself to create a somewhat decent looking UI was not something that came naturally to me as a data scientist/data engineer. I am not a web developer so at best, my output was clunky and not very pretty. With Cobuild, the bar has been significantly lowered in being able to produce professional looking apps with minimal effort. Shew.
- Documentation. With our clients, we advise them to create project documentation inside the wiki of each project covering inputs, outputs, and descriptions of each flow zone. While everyone agrees this is a best practice, few people enjoy doing it. Using Cobuild, I can simply ask it to produce this documentation. Even prior to Cobuild, you could export a Dataiku project and ask an AI tool to summarize it. Now, I don’t even have to copy/paste data outside of the platform, risking security breaches. It’s such an incredible benefit and saves time doing tasks that people don’t enjoy doing.
- Check projects for compliance to best-practice checklists. This is one of my favorites. In our work, we assist clients in building up their Centers of Excellence (COEs) around analytics, many times with Dataiku at the core. In this work, we establish checklists that builders need to follow for their projects to be deployed to Automation. This is to ensure robust CI/CD and create consistency and reliability of the data products coming out of Dataiku. To facilitate checking the project against the checklist in the past, we used Dataiku log files and created a Dataiku data pipeline with outputs visualized in Power BI. This included things such as: Snowflake (or database of choice) dataset names should match the Dataiku dataset names, use the SQL database engine as often as possible, make sure the project description and wiki is complete, make sure there is a scenario with a trigger and reporter and no hard-coded variables. Then Cobuild came along. I simply pass my checklist to Cobuild and prompt it with, “Here are my best practice checklist items. Please check the project for adherence to these.” Then in a few minutes, I have a full diagnostic. All I can say is, “Wow.”
- Whiteboarding and brainstorming. Sometimes we approach our data with a predisposed idea of what’s possible and exactly what we need or want. I see Cobuild as a potential collaborator for brainstorming. You can start with, I have this data, I have this idea, what could we do with this? This could be in the range of descriptive reporting, predictive analysis, and generative AI uses. It opens the door to the exploration of new uses of data we already have. Neato.
- Identifying what processor or recipe I should be using. This is probably my biggest use of Cobuild right now. In a Prepare recipe there are literally 100 processors (functions) to choose from and sometimes I simply forget which one I need. For example, with pivoting and unpivoting, there are folds, unfolds, pivots. I can’t remember which one does exactly what. So rather than spin my wheels trying them out, I simply ask Cobuild, “I’m needing to do this. Which processor is that?” The same applies to recipes. There are many visual recipes with overlap in their capabilities, but there is often one best choice. Cobuild can save you time and direct you to this best choice.
In all these cases, I am asking AI to do work I already know how to do. It is saving me time and sometimes producing a better product by doing the work I don’t like to do on my own.
Given these incredible uses of Cobuild, when would I not want to use it? That’s easy for me to say. I use the analogy of driving. Self-driving technology is amazing, just as Cobuild is amazing. Autonomous driving can work well under certain conditions, such as clearly marked roads and good lighting, but it isn’t always going to work well.
There’s another important point. I like driving. I’d rather drive than be driven somewhere, especially if I still have to navigate. The same applies to work I do in Dataiku. When I already know what I’m doing, I’d rather do it myself. It is faster and more satisfying, not to mention less expensive. If I ask Cobuild to build my data pipeline, I have to explain it well and then check its work. For me, this takes the joy out of using Dataiku.
In conclusion
I am very happy to have Cobuild in the Dataiku platform. I do think it has forever changed how I work in Dataiku. It has made things I don’t enjoy doing, like building a webapp or creating documentation, so much better. And it has streamlined best practice checking and become a great collaborator.
I do not think I will ever abandon the traditional way of working in Dataiku as some have claimed. Just like I want to drive myself to the mountains this weekend, I want to stay in the driver’s seat as I build out data pipelines with Dataiku.
Amplifi is here to assist organizations in right-sizing the use of AI inside Dataiku and in other tools and platforms. With our deep experience with data management, data engineering, analytics, data science, and implementation, we know what it takes to deliver accurate insights in a sustainable way. Knowing where and when it’s appropriate to plug in AI agents and tools like Cobuild is essential in preserving your investment in your data platform. Let us help!
About the author
Shelly Kunkle is an Analytics & Data Science Practice Lead at Amplifi, where she helps organizations build scalable data pipelines and Analytics Centers of Excellence using Snowflake, Dataiku and Power BI. She joined Amplifi through its acquisition of Snow Fox Data, where she was a Senior Data Consultant. Before consulting, Shelly spent several years at Michelin in a range of analytics roles, most recently as Global Lead, Customer Data Analytics, and she later founded her own analytics consultancy, Every Stone Data. She holds an MS in Analytics from Georgia Tech.
