Newsletter #19: 5 Steps for Building Machine Learning Models for Business
Featured posts from Slack, Pinterest, and More
Hey there,
As always, you can read my top discoveries for the last month below. If you like this newsletter, please forward it to others who might find it useful.
Better coordination, or better software?
How do we make coordination between teams easier appears to be a reasonable question to ask. But by making coordination easier, we enable more of it. That means more coupling between teams, and the more dependent teams are on one another, the more costly it is to manage cross-cutting concerns. Instead of looking for ways to improve coordination between teams, how about doing less of it? One of my major key takeaways from this post is to minimize coordination by establishing boundaries and the few interfaces that cross teams to reduce multiple teams' coordination.
How Slack Design their APIs
I spoke at a tech conference a few months ago and where I said the secret to building APIs that developers love to use to make them simple and easy to use. Making APIs dead-simple to use is paramount to driving adoption. This is why this post resonates with me a lot. I believe slack’s principles to building APIs are important for any team building an API product.
5 Steps for Building Machine Learning Models for Business
We often think machine learning is the solution to all problems. That’s far from the truth. Not all problems are machine learning problems. It's important to ask if it's the right time to build a machine learning model especially if you're building a new product. Can you start with a baseline solution and improve it down the line with a machine learning model when needed? This post contains 5 steps for building ML models for business.
How we scaled the size of Pinterest’s ad corpus by 60x
As Pinterest's ads corpus grew, memory bottlenecks began to surface. Instead of scaling the existing service horizontally or vertically, the ads team moved the in-memory index to an external data store. In this post, Nishant Roy shared their journey and how they leveraged a key-value (KV) store and implemented some memory optimizations in Go that helped them scale their ad corpus by 60x.
Cheers 🎉,
James