2025-08-01
I’ve been working, learning and building. Tons of it.
I also got a permanent contract at my job after my 6 month probation period ended, so yay!
I find myself lately working more, not less. And I love it.
This happened because knowledge is readily available and the only scarce resource is now time. Of course there are cases where AI underperforms, but in general you can build almost anything.
The Jevons Paradox is happening in real time and if this trend will continue, which I think will do, the whole society will finally benefit from it.
Enough now with the AI hype, and let’s talk work.
I love how indie hackers such as Pieter and Marc build in public and make a living with it.
Another thing happens with AI is that there is an overload of applications but you will pay only from the people you trust and follow.
What it does: dashboard with data regarding my daily route.
Why build it?
I want to see when it’s the best time to drive into office, how fuel prices affect traffic, etc.
As mentioned in this blog post I’ve created a dotNet backend to gather information about daily route since March. I will now analyze the data, create a dashboard and try some of the AI Agents Frameworks like ML Intern or Andrej Karpathy’s Autoresearch to also train an ML model. Building a pipeline to trigger the training if some metrics fall below a certain value might also be part of the project.
What it does: finds the best offers for keto products (meat, eggs, cheese, etc) in a radius of 5 km from my postal code.
Why build it?
Because I need it of course. Why not find the highest quality products in a radius of 5 km from where I live, and never have to ask myself if I got the best deal? At first I will build it to scan only for supermarkets but extend it with a ‘local’ option where you can find small farms and also get their products.
The user interface might be tricky, but it’s the perfect use-case to solve a problem I face at least once a week.
What it does: evaluates your AI agents with metrics which are relevant to you.
Why build it?
Because as far as I know it’s the first product which allows you to evaluate, spot regression across your system and allow you to use your own metrics.
Evaluations in general are hard, but necessary. A big shoutout to ragas.io for building a great product.
One of the biggest problems when developing agents is that the system is complex and is nearly impossible to spot a regression when introducing a change.
This motivated me to build a system that not only evaluates their response, but also find regressions in the performance.
The workflow runs like this:
The biggest challenge is to define the ‘golden’ dataset of questions and answers.
The design of the tool also allows it to be integrated in all types of applications. This is a topic for a new blog post.
What it does: keeps your mind sharp and grounded in reality.
Why do it?
To be a good engineer you need to do more than just build.
The amount of information I have to process every day feels like drinking from a firefighter’s waterhose. Jokes aside, I need some activities where I don’t use AI but at the same time I sharpen my technical skills.
So I’ve learned:
Neovim: I force myself to use Neovim as a text editor, including for writing blog posts. Remembering all those shortcuts keeps the mind fresh.
Code without AI: I just wrote the code for the attention mechanism from the transformer by hand. Again, keeps the mind sharp.
Writing: I keep writing this blog, without AI.
Record a journal: started small, and started recording myself whenever I do a walk alone, to improve my speech ability.
The last one is a must for me, because I really enjoy creating, also content.
So let’s see where this journey will take me.