Another day, another standup, another new JavaScript library, another broken agent-based framework, another non-existing dataset, another failed training experiment. This is the world we live in. At times this industry feels like a complete mess, other times you hit your stride and feel like a problem-solving guru. We are going to learn how to become better at our skills as we navigate this crazy world.

It’s July of 2025 and the AI slop is raining down on us. It’s time to level up together. It’s time to hold ourselves to a higher standard and conquer this arena. The goal of this blog or technical discussions are to explore and deepen our technical knowledge of AI/ML and Software. I have been working in the field of ML and Software for a while now (~7 years). I have been in charge of multiple projects from ML research all the way to production software. I am a weirdo who likes all sorts of problem solving and will work on ML, backend, frontend, or anything needed to get the project done.

ML, do you mean if/else?

Machine Learning is not a simple problem space where you can make accurate estimations on how long a task is going to take and if it will be successful. I don’t know if this dataset and algorithm are going to work but I have a good feeling it will so let me try because the current model is awful. I will come back in a week with a report. Do not ask me to point this for our sprint velocity. Please do not ask me if an LLM can solve an object detection problem in a niche domain.

No you can’t use an LLM to solve every problem on this earth. Yes ML is hard and you can’t expect to vibe your way through it. On the other side of this equation are all the ML nerds who can’t ever get their algorithms/models to work outside of a jupyter notebook. They have no idea how to track experiments and at the end of the day they hand you crap that will never solve the real problem. Now someone else has to take this steaming pile of crap and compile, optimize, rework it until it’s a somewhat mature solution that won’t break the moment real data hits it. We will learn how to research, train, experiment, and optimize models for production environments.

Software Woes

This is a wild time to get into software. All these VCs and suits keep yapping about vibe coding the next big thing. Meanwhile I am reviewing MRs where the vibe slop is infesting the project to a point where this MR is broken and the person who copied this code also sucks for making me attempt to understand the context and how this is even supposed to work. But it’s ok because the crappy generated unit tests pass and are green…

We have to grow and learn how to make better software and products. If you are really into ML but are weak in software then we can grow and get better together.

Communication Issues

I see people struggle or solve problems in ways that set their co-workers behind even if it’s not intentional. That code you just copied from an LLM only solved the happy path and it changed the API. None of this was tested well and the docs were not updated so now Bob has to work with the mess you made. I am not a strict clean code guy (I think many topics in that book are taken too far) but you need to solve problems in ways that help your co-workers, management, and shareholders to make their lives easier.

When presenting your work to your peers, management, customers, etc… you need to know how to talk at different levels. Your fellow devs understand that you can’t spawn threads in an infinite while loop. The CEO does not understand why you keep yapping about how upgrading node modules keeps breaking things. These are all things we can improve on together.

Standups are depressing and please don’t talk about your stupid dog or how you ordered the wrong pizza. I got work to do and don’t need to waste time on this insanity that the industry participates in.

High Level Goals

  • Learn how to create ML models and algorithms
  • How to analyze data
  • How to compile models and different optimization strategies
  • Become better at problem solving as a developer
  • Communicate more effectively
  • Marry ML and Software together
  • Learn how to use distributed computing for training models
  • How to best serve models based on project requirements
  • How to structure teams for better success
  • Rant about various topics

I look forward to having you all along this journey as we crash and burn our way through CUDA versions and SEGFAULTS.