Upskilling technology and culture

Posted by Scott Speights    on May 3, 2019 in Dev tagged with Culture, Data, Development

Upskilling as motivation to transform

It was a surprise when our CTO Raul told me half a year ago, that he wanted to upskill a tenth of our software engineers in the basics of machine learning and artificial intelligence, and, that he wanted me to both organize and take the training, so I could start doing ML too. It was a good idea and a great opportunity to expand the different types of technology events we offer to support company technology and collaboration targets.

The topic of machine learning was a good selection and made a lot of sense. On the one hand ML provides us a completely new way to create products and features by leveraging ever-more-powerful, constantly-developing technology, on the other, ML has the capacity to free up resources, to streamline work and to enable engineering data-centric solutions for people, who don’t have specific domain knowledge in math and statistics. Maybe this is why machine learning has been added as a major building block to our technology strategy and why mastering machine learning is seen as a competitive must by top management.

Upskilling Conways law

Our previous CTO, Holger Reinhardt, was enamored with Conway’s law and maybe even a little obsessed with it. Part of his work was to persuade us to learn the right technologies, so we could eventually build “microservice-like products” in a not-too-distant future. His mantra was “technology, structure and culture”, and he never mentioned it, but inverse Conway maneuver, which you can find as a regular blip in the ThoughtWorks technology radar, was also part of the plan.

ICM sounds like a professional wrestling move but, basically, it’s just taking steps to transform teams and organizations to produce the types products you want to produce. This involves a lot of time and a lot of work because inevitably changing an organization also requires changing communication patterns, culture and mindset, which is not a simple task: if you believe in Conway’s law then you must change the communication structures in your organization (or add new ones) to be able to create new kinds of products.

The ML upskilling has been one great tool to provide impulses to inverse Conway the technology organization at Haufe Group: the upskilling has been effective for enacting organizational transformation because more people gain basic skills to take back to their product team, and, the people in the training get to know each other.

Learning to engineer ML solutions

For the upskilling itself, we created the program together with the AWS professional services data science team. The training, all in all 10 days, covers the basics of ML for topics like what you have to know to do ML, the workflow for solving ML problems, and which kind of problems can be solved with ML. The training was delivered by AWS Data Scientist, Will Mcgehee.

Over three sessions, we covered a variety of supervised and unsupervised machine learning tasks based on our own Haufe Group data and own use cases. The tasks ranged from using community filtering to recommend articles to customers looking for Haufe Group content, to using clustering techniques for creating customer segments, to making use of modern NLP strategies to search an entire corpus of content for relevant matches.

By solving these problems hands on, we got an end-to-end look at machine learning, not only training and evaluating models but also cleaning data, selecting features, and interpreting visuals to see how good our results were.

Learning to develop ML products

Somewhere along the way, we realized that it would be important to not only skill up our engineering staff, but also to get knowledge about “doing ML” to people responsible for managing and developing Haufe Group products. Here we were very lucky to find out that our partner, AWS, could also provide workshops for working with ML for product development. And so, we added this workshop to the AI Foundations.

It was important to activate the product managers and make them sensitive to ML. Doing so has given our product teams a shared understanding of ML, which has spurred development of actual ML features and increased our chances actually doing something with this technology.

The teachable moment

The upskilling has been successful on several levels. First, the people who took the course, including yours truly, captured the ML basics by working on Haufe Group problems and were then able to go back and start experimenting with ML for their own products in parallel with the training. That’s what many of the course participants did. Check out our work with German Recommendation Letters Part 1 and Part 2. ML features and product are starting to emerge from the product teams. We might not be able to use ML for everything but we will be able to leverage it where it makes sense.

The 40 engineers, who have taken part in the training, now form the basis of the ML Community here at Haufe Group – a go to target for asking questions about and getting answers for Machine Learning at Haufe Group.

We’ve had a great co-creation experience with AWS: Big thanks to AWS proserve and to my colleagues at Haufe Group for the hard preparation and great working together.

Finally, the folks, who took the course, are more highly engaged than ever to be working for Haufe Group. We’re still on the road, as far as mastering machine learning and transforming the organization goes. Investing in our people so they can accomplish more is a great way to catalyze these transformations.

Stay tuned for more news about upskilling in HG Dev Community and how it impacts our culture. And, check out Will’s take on the upskilling next week on the dev blog.

US Navy Blue Angels, Double Farvel maneuver CC BY 2.0