kar0lina

I'm the one who made the project featured in https://news.ycombinator.com/item?id=48704289

Incoming 4th-year CS major @ GT, looking for contract work (~20 hours a week) from May 2026 - Aug 2027. Especially interested in robotics, RL / embodied ML opportunities; security engineering background. Internships: 2x Apple, SpaceX, Samsung

Location: Atlanta, GA, USA

Remote: Yes

Willing to relocate: Yes, after grad. Need to be in ATL for the school year

Technologies: Python, PyTorch, MuJoCo, PufferLib, reinforcement learning (PPO), C#, Java, JavaScript/TypeScript, SQL, Linux, Kubernetes, Wireshark

Résumé/CV: https://karolina.mgdubiel.com and https://linkedin.com/in/karolinadubiel -- please email for actual resume :)

Email: karolina+hn [at] mgdubiel (dot) com

Hi, thank you!! There's so many great resources for building your own quad from scratch (ex https://www.youtube.com/watch?v=S9dvDU5RmQU&t=476s). I honestly wouldn't recommend an octo unless you have a reason for it (like wanting more fault tolerance, etc) solely because it's twice as expensive for the motors. I got all my parts from GetFPV and Amazon. At the very bare level, all you need is a FC, ESC, frame, props, motors, receiver, and battery (plus controller, and goggles/camera if you want it to be FPV)

Hi, I’m Karolina! I get it!! I’d probably have some doubts if I saw this online too.

To be fair, I definitely did not get good at Fusion in a day (I’m still not very good at Fusion). The Day 1 CAD was all flat shapes -- all I had to do was use the line, curve, and pattern/mirror tools and then extrude those shapes to my body plate / arm thickness. Same with the screw holes -- I didn’t know there was a hole tool and was just drawing circles, patterning them, and extruding them backwards. It still took me forever, I started super early in the morning and worked on it until ~2pm, knowing it would probably take a MechE major 30 mins to do the same thing.

I also do have mentors in the form of eng majors who can give me verbal feedback / advice / show me which tools to use / get me unstuck. To me, it was very important that I was the only one actually doing the CAD so I could learn, but I definitely have had super amazing mentorship along the way to teach me and point me in the right direction. The motors and props are just 3rd-party components that I imported (still took me ages to figure out how to constrain them right). I went to the machine shop after and that took forever too, especially figuring out how I’m supposed to arrange the plates for printing in Fusion (but I had people to help me out there, they were not going to let me use a CNC mill alone with no exp lol)

I made the GPS / receiver mount way later in the process because I was so intimidated by the CAD for it (literally taped my GPS and my receiver to the drone for the first flight instead), and it took me days (also such a simple component, but the stuff that’s not just an extruded 2D shape is way harder for me).

I’m definitely not superhuman and it’s probably going to be years before I feel truly confident in Fusion :))

Newest is at the top :D If you press the little "Collapse All" button on the top right of the blog posts, you can see the full timeline + dates of the entries.

I did start by designing/building a regular FC (no RL or anything) and am doing RL training afterwards, so I agree that the whole timeline looks a little confusing without background context since it goes real drone --> sim haha

Hey! This is a great question because RL is overkill for my use case, you're right! The reason I'm using RL is because I wanted an RL project on real hardware, so I built this idea of a drone project around that concept, not vice versa. A lot of people on X have brought up alternative solutions -- like MPC -- that might be a better approach to this problem engineering-wise. I just really wanted a real-world RL problem and this was kind of the first cool idea I thought of :D

I'm sorry for the late response, I didn't realize someone had posted my project

I use Claude Code to help me with the updates on my website/blog because I have a full time job and try to squeeze this side project into as little time as possible :) It's definitely not fully AI-generated, though, I love writing, and I'd consider myself "savvy" enough haha. But it's ok if you don't agree! Sometimes it's easier to use Claude to condense ideas (for ex, the abstract) when I'm on a time crunch. But either way, Claude definitely didn't solder, CAD, or CNC-cut anything for me :D

I don't think there's any shame in using LLMs to save time on documenting your projects. Let's be kind and positive to each other!!

Hi! Thanks for your kind words. I actually had no idea this got posted here, I'm Karolina, and this is my project! Really cool to see some feedback here! I went with CNC cutting over 3D printing just because I wanted to avoid any stiffness/vibrational issues. I don't actually have any prior drone experience, so I wanted to play it really safe with that.

^^ This was why I opted not to 3D print the whole thing :)

Hi!! Would love to connect :) My X is karolina_dubiel and LinkedIn: https://linkedin.com/in/karolinadubiel

Hi!! I'm Karolina, and this is my project! Had no idea this got posted until someone DMed me about it on LinkedIn. Feel free to ask any questions or follow along via X (karolina_dubiel)

Thanks so much for all the kind words, it means a lot!