Recent content in Home on Handmade Oasis
Three months since the
 last update , give or take a day.
In that post I landed on the idea that the layer around the model is where the
leverage is, the harness, the context, the review loops, all the boring pieces.
I still think that’s true. What changed this quarter is which problem I point
all of that at. 
 Because what I have actually been optimizing for is not…
As I was writing this post I could not crystallize who the target audience for
it is, other than for myself to put into writing how I view and think about
LLMs. I have spent enough time with these systems by now that I have stopped
trying to map them onto things I already understand, because every analogy I
reach for ends up breaking somewhere. The most honest thing I can say after…
It’s been three months almost exactly since I
 last documented my workflows
and setup. Although it does not feel like it for me, reading back what I wrote
in the last update there are quite a few differences. 
 The system I have going now really does feel like an exoskeleton, to quote a
recent interview with DHH where he described it like having 12 arms and a
totally…
Last time I wrote about my AI assisted workflows was 3 months ago. Paradoxically,
everything has changed while nothing feels different. Many of the techniques and
methods I was using 3 months ago were new to me. As I described in
 that post , it felt like a
Frankenstein type of
workflow. Many different systems duct taped together into something. That
something I, in…
Alright, we covered the conceptual stuff in Part 1a. Now it’s time to actually
look at some code. In this post I’m going to walk through the entire agent.py 
file line by line. By the end you’ll see that there really is no magic here,
just a while loop, some API calls, and tool execution. Let’s dig in. 
 Getting Started 
 Before we dive into the code,…
Before we get into the coding part of the series I think it would be beneficial
to give some background to what the ReAct loop is that we are going to make use
of to build this agent and also describing a bit how the code is going to look
like. This will not be that long to be fair because the basic loop is fairly
simple once you grok the ReAct loop.
Ever since I switched to Linux as my main OS and started building my own
tools, I’ve had this nagging feeling about the coding agents I use daily. I
understand the principles how LLMs work, how tool calling happens, the
general architecture. But there’s a difference between understanding the
concepts and truly knowing what’s happening in the implementation. 
…
Current limitation as I see it 
 So I have been experimenting quite a bit with LLMs recently (see my other posts,
especially /state-of-ai-assisted-workflows-october-2025/ ) and dedicated some
serious amount of effort to play around with the tech to both learn more but
also to build an intuition for what it is and what it can do well etc. This has
led me to having a realisation…
This space is moving at a dizzying pace currently and as such there are
many new “frameworks”, methods and approaches popping up and promising the
world. But having spent the past 2 months trying out many of these on
non-trivial tasks I have managed to nail a workflow which I have pretty
good success rate with. This post is intended to be like a context
compaction…
I wrote a post a while ago about my current workflow when working with AI , my
initial skepticism and some brief mentions of the concept of recreational
programming. This post is going to be a continuation of the train of thought
presented in that previous post, but it will be a stand alone piece still. In
other words there is no need to have read the previous one to follow along…
Lately I have been thinking quite a bit about the question “What is a data
platform in its most basic form”. When I set out to answer this question my
first thought was that it’s not possible to answer this in a useful way. Reason
being that what a data platform is varies a lot depending on the company.
Basically, the data platform that a big tech company with…
This is just like a note or dump of everything I research on AI agents, MCP, and
other generative AI stuff. I think the correct phrase for it is a stream of
consciousness dump 
 There seems to be a kind of tier or ladder of sophistication that is slowly
getting shaped when it comes to AI assisted workflows, especially in relation to
software engineering. As I see it currently…
Let me preface this post with the fact that
I was a huge macOS fanboy for a decade or so and only started feeling that the
OS was getting in my way a couple of years ago. 
 With today being the 34th birthday of Linux it made me reflect a bit on how I
have used it over the years. I started out using Ubuntu at University 12 years
ago for some intro labs and for learning how to…
About Me 
 Hi, I’m Ramtin Javanmardi! 
 Welcome to my corner of the internet where I share my thoughts and experiences on: 
 
 Data Engineering 
 Developer Tooling 
 Low level programming 
 Game and graphics programming 
 General Technology 
 
 My guiding principle when it comes to engineering is: Always aim for great, not
good enough, not good and…