Listening notes #1: Montana Low, founder of PostgresML, and Swyx on running ML models in the database

DanielMarch 22, 2023March 24, 2023

I am an avid podcast listerner. I think the biggest amount of information I consume is by listening to stuff. For me, nothing beats the feeling of a long walk in nature while listening to some interesting talks.

I have always tried to find a way to blend the listening hobby with this blog. And inspired by the reading notes of Jim Nielsen decided to create a new type of post entitled Listening Notes. I know .. very creative 💡!

So, I will start this new content category with a great conversation I listened to this Sunday morning between Shawn Wang aka Swyx and Montana Low the founder of PostgresML:

Some of my main takeaways from the conversation:

• ⏱ 00:40: machine learning is more about data, that is about algorithms.
• ⏱ 04:40: given they need a super fast way to detect if an order is a fraud or not Instacart uses AI and Machine Learning to detect fraud payments almost real-time.
• ⏱ 08:10: the core concept of PostgresML is to bring the model to the data, (vs the classic approach that brings the data to the model), thus increasing the response speed.
• ⏱ 09:35 for deep learning models the maximum memory requirements are 4x to 14x times the number of parameters. You can easily end up with 2.5 terabytes of data.
• ⏱ 50:45 structuring ML problems is really hard. Taking a business problem (like is this payment a fraud) and turning that into an ML problem that can be solved with an ML model is a skill set that not many people have
• a great conversation about how to do a practical DevRel setup for a startup. Regarding this subject also this article is quite nice: Measuring Developer Relations

Keep learning and movin' on!