I’ve failed a lot of interviews at different stages, even after pulling from a crazy referral network.
I’ll keep applying of course, but this is suggesting that my skill level is not at where it needs to be.
I’ve spent 2026 waffling around. I didn’t make much progress at my career, and I haven’t made much additional progress w.r.t. philosophy, math, CS, really a lot of things. Somehow, I feel more competent than I was in 2025 in a lot of ways, but I can’t remember any “real defining” thing that I did.
Somewhat Concrete Projects
We’ll have “somewhat concrete projects” written down here.
Distributed LLM Serving/Training
For some reason, throughout all my time of learning about LLMs, I have never touched this aspect of LLMs. Back when I was in college (well, new grad but unemployed), I studied the CUDA programming model, ThunderKittens, and esoteric things such as CuTE and CUDA custom hardware primitives.
I think this highlights one of my common failure modes - in studying something, I often spike in some random area but then forget the basics, because I don’t want to start all the way from the beginning. Like, do I want to know arcane CUDA primitives, or do I want to learn about LLMs? Those are two different questions.
But most likely, the core constraint is cost - multi GPU deployments are expensive and hard. I’m still thinking that will be the core issue with this.
I’m eyeing the CS336 Stanford course. If not that, then I will sprawl through MOOCs and try to learn about these things.
Time is ticking down fast, though, I feel. Maybe this is unwarranted, but it feels like the world is moving too fast and I just can’t keep up.
Linux Kernel Virtualization
Still on this topic. I think kernel programming is cool and I want to actually get into something related to linux and hardware.
Part of it is admittedly kind of aesthetic - I’ve heard of
kernel maintainers getting into the zone and using tools like
lei and b4 to consume and build patches
at an insanely fast rate. Maybe a completely arcane and obsolete
tool, but still cool.
On the whole though, this probably should take a back seat to other endeavors.
BTD Battles Network Architecture
This one’s been cooking up for years now. I probably don’t want to make the game itself in its entirety (though maybe I should - that would just practice a lot of “mundane” software engineering skills that I nevertheless am not learning on the job right now). I want to at least formalize the protocol.
The protocol is super basic - lockstep with provable interaction boundaries in between, so the clients can be mostly authoritative with little latency while things like desync should (in theory) go away - but there are interesting extensions such as, how to actually test this kind of distributed architecture in reality.
Bonuses
Random bonus projects that are big enough in my life at this point to make note, but not nearly concrete enough to do anything with.
Bonus: Things Fourier, LTI, linear algebra, diffeq related
It’s still technical but this is a pretty VAPID goal, so it’s added as a “bonus”.
I didn’t take any applied math or pure math courses. So all of this is super new stuff to me. As I’m doing more RF work, I found out that you study circuits with the Fourier Transform (and often a more generalized version called the Laplace Transform), and that terms like “bandpass, lowpass, highpass” filters all are just normal terms in circuit design (no RF context needed). Like, wow.
I did competitive programming and studied some TCS on my own time. I’m very much an armchair kind of guy - “by construction” is one of my favorite ideas, whereas the physical world often yells at you for such elegant phrases - it will kick you in the ass if you don’t accept its constraints.
I have no clue how I’ll get actually better at the high level concepts here besides exercises and reading, but I will be expending non-trivial effort in this area.
Differential Equations and Linear Algebra
Signals and Systems
here I come!
Bonus: OpenTelemetry, observability
(Added 9-08-2026)
Not the specific framework, but the ideas behind observability and automating “scientific” models in some sense w.r.t. a software system are becoming more and more pronounced.
This and the math are directly related to my current job. I think I want to not do too much unpaid labor, but this kind of work will directly leak into my other projects regardless.
Bonus: Philosophy?
I had my philosophy (of science) arc - very interesting, but I’m unable to fold them into actually good observations. Philosophy of science is interesting, but as I said before, analyzing the problems with doing the thing is not the same as doing the thing. It can help - greatly! - but you still need to do the thing, you know?
I’ve mainly been reading philosophy because it really isn’t that crazy. Turns out that the field that thinks about thinking, has thought about thinking really, really hard, and when people from all kinds of fields try to understand “thinking about thinking”, there are common (at least, human) patterns there.
I still have a ton of analytic and especially continental philosophy on the backburner that needs to be consumed. I am making an executive decision to put it on hold - I am stagnating in my career among other things, and I need to get my life together first.
The one exception is Making It Explicit. Inferentialism is my favorite LARP term, and I always use half-baked memes from it. The issue is that it draws from Kant, Hegel, and a lot of analytic philosophy - so all the background pieces I’ve been meaning to read, but haven’t. I’m down to just read it end to end with a bunch of misunderstandings, just to actually have read it once.