About PUMKIIN.TECH
The Road to Senta (Paved With Failures)
I've always found AI systems cool. Wanted to work with them, play with them, build them. So I tried. And failed. A lot.
First project: build my own LLM. I followed Andrej Karpathy's tutorial and it felt cool. Downloaded all of English Wikipedia as training data because why not go big? My 3070 on the otherhand said no. Somewhere between maxing out its 8GB of VRAM and using all 64GB of my system memory I realized. It didn't work at all.
Moved on.Some time later I made a branching tree ML model in a beginner Python class is was bored with. It identified suspicious network traffic vs. normal with 98% accuracy. While everyone else played with turtles drawing shapes, I built that. Then I dropped out of college immediately after because it felt too easy, too expensive, and not about what I wanted.
I tried teaching a neural net to play Wordle using reinforcement learning. That failed horribly. Who have thought teaching a nerual net to spell english words, where the word changes everytime is difficult. Apparently I didn't.
After all these failures, I realized something important: I had no fucking clue how any of this worked.
So I hit rock bottom and started small. Built a simple sentiment analysis bot. Positive/negative at first. Then moved on toemotions. But I needed training data, so I made a demo page for me and my boyfriend to label sentences. One night at his workplace, I floated the idea: what if I turned this into a full website? A tech portfolio to show what I can do. Not to get hired, just to prove I could build it. To talk about it for anyone to read.
So I built a proof-of-concept homepage in about an hour. That's what you're looking at now (with some polish, of course).
Senta actually worked. And PUMKIIN.TECH was born.
How I Got Here (The $40 Origin Story)
I've been around tech my whole life. My dad loves computers, taught me a lot. But around age 6 or 7, I started tinkering on my own. By 12, I'd built my first gaming PC from parts my dad bought for christmas.
At first I ran Minecraft servers, played with network shares, tinkered endlessly. Eventually bought an old Dell Optiplex 9020 SFF for $40. A single 4TB drive, amd it ran Windows at first, then Ubuntu. Docker containers. Reverse proxy. Jellyfin. Digitized all my movies on DVD, Blu-ray, and VHS.
At some point I scaled up—bought a PC case with 12 drive bays, filled it with 4TB drives, installed TrueNAS Scale. Around age 15-16, had a falling out with my mom. I ended up giving up on tech for a while. I worked at Walmart pushing carts. The upside was I lost some weight, but I also lost myself.
Eventually started buying things I wanted again. Got a UDM Pro from Unifi, some WiFi access points, a PoE switch. Built my own network from scratch. VLANed my part and my dad's part separately. VLANed web services into their own segment. I started thinking about security.
I found an old Supermicro machine on eBay. Built it up with cheap hardware. And it still runs today. Proxmox, 128GB RAM, 12TB of SAS drives (which I want to replace with SSDs), dual Xeon E5-2690 v2 CPUs (10 cores, 20 threads each). It runs everything. Sits a couple feet from my bed, humming day and night. Critical stuff runs on Linode VPS for uptime. Everything else? That Supermicro box in my trailer.
What to Expect Here
I write about whatever inspires me. Right now that's Senta (the sentiment bot that finally worked), image processing fundamentals, and whatever else catches my attention.
I'm not a how-to guide. I'm not smart, I'm not friendly, and I'm not always nice. But if you're here, you probably don't need another polished tutorial anyway.
This site runs on what I built. My Proxmox homelab (that Supermicro beast), Linode VPS for uptime-critical stuff, PHP, Python, MySQL, Redis. No frameworks. All built from scratch.
You don't need what I have. My homelab started with a $40 office machine from eBay. I never thought I'd get to this point. I live in a trailer where power draw is a real concern. There's still a lot more I want, but I'm already pushing the limits of what these walls can handle.
Start small. Fail a lot. But build anyway.