Some notes and takeaways from the Devit 2018 conference that took place in Thessaloniki on Monday, June the 11th, 2018 and that will also help my long-term memory.

From David Platt

Lively presentation!

Reading:

• The joy of ux

From Pawel Dudek:

"There is no such thing as untestable behavior". If you cannot test your code then you haven't architect it correctly. "Tests drive the architecture of the app".

To checkout:

From Cheryl Platz:

Ask difficult questions when you design and build apps in the era of AI and conversational bots like:

• How will this make the world better or worse?
• If we are successful how customers will be harmed?
• How customers can abuse our product?
• What is the worst case impact our product could have?

To checkout:

• The tarot cards of tech for humanity-centered design
• On Surveys for UI/UX
• The practical ethnography book
• Microsoft's Inclusive Design thinking
• dscout

Small note: Device agnostic architecture - move customer data and telemetry on the cloud, don't leave them in the device. So the user could have a nice flow when moving between devices.

In general the theme from both David Platt's and Cheryl Platz's talks was similar to the mom test book:

• "Don’t build products your customers don’t need"
• "Give customers what they want for a price they are willing to pay"
• Don't write code before showing mockups to users/customers (leads to expensive, lost time)

From the panel on privacy

I learned that according to research a person has to spent 76 work days per year to read all the Privacy Policies encountered on the interent.

From Ingrid Epure

Notes:

• Add thresholds in dashboards so the user does not have to think so hard about a metric

Checkout:

• How complex systems fail paper
• http://opentracing.io/
• https://www.honeycomb.io/
• Tom Wilkie's RED method on the metrics you need to monitor

From Julien Simon

Fun demo with a small robot on how one can use AWS and move all computations to the cloud with various AWS services.

I am thinking that from now on it doesn't really makes sense to build and deploy your own models that for example do face recognition or text translation. It doesn't worth the time and effort.

Also checkout:

• Two Decades of Recommender Systems at Amazon.com paper