alott@umbra · loc: [redacted]

alott

"I build things because I want to know if I can."

DeveloperAI / ML CybersecurityBuilder FishermanEntrepreneur
// about 0x01

Self-taught, and mostly awake at 2 AM.

I'm a self-taught full-stack developer who's into AI/ML, cybersecurity, cryptography, and building systems that probably shouldn't work on a first try — and then making them work anyway.

I spend my time experimenting with software, training my own models, wiring up AI agents, and figuring out how computers actually behave under the hood instead of just reading about it.

Outside the terminal I'm heavily into fishing, music, boats, and anything that involves taking a thing apart to see what makes it tick. I especially like combining tech with the stuff I enjoy — like using AI and computer vision to find fish.

Still early in the journey. The plan is to keep building increasingly ambitious things and turn some of them into real products.

Grade
10th
+ AP Computer Science Principles
Primary language
Python
full-stack · CUDA-adjacent
End goal
$1B
a SaaS company, eventually
Method
Hands-on
learn by breaking, then fixing
~/stack 0x02

What I actually work in

Core

PythonFull-stack web AlgorithmsGit / GitHub LinuxOperating systems

AI / ML

Neural networksLLMs PyTorchCUDA GPU computingModel training

Security

CybersecurityCryptography PentestingRecon Web probingVuln reporting

Systems

VMsContainers AI agentsVRAM tuning Batch scalingToy curves
~/rig 0x03

My setup

alott@umbra: ~/rig
alott@umbra:~$ neofetch
alott@umbra:~$
~/projects 0x04

Things I've built

flagship

Umbra OS

an AI-controlled OS / VM environment for autonomous agents

A virtual operating-system environment built around AI agents that carry out programming and security work on their own — the OS is the body, the agents are the brain. Point it at a target and it goes hunting.

Programming work Security auditing Pentesting Network recon Web probing Directory scanning Vulnerability discovery Automated reporting Agent orchestration
▸ view on GitHub — alott2223
~/research 0x05

Experiments & rabbit holes

Neural nets, from scratch

Not just calling an API — actually training GPT-style models on my own GPU.
Trained GPT-style models at ~150M and ~21M params
Dataset: WikiText-2
PyTorch + CUDA / GPU acceleration
Batch-size sweeps & VRAM-utilization tuning
Training-loop optimization for a single 4090

Cryptography, hands-on

Poking at the math behind Bitcoin to see what actually leaks.
Elliptic curves, finite fields, point multiplication
Discrete logarithms & the Pollard kangaroo method
Brute-force / key-search experiments on toy curves
Tested whether probes leak info about a secret scalar
Bitcoin-style cryptography, taken apart
[ sonar ] 0x06

…and then we leave the server room and walk into a swamp.

Personal bestSpeciesNotes
46"Northern pikethe one that made the net feel small
39"Muskiefish of ten-thousand casts, caught
37"Northern pikebackup proof the 46 wasn't luck
6 lbSmallmouth bassa genuine tank of a smallie

On the board

LargemouthSmallmouthNorthern pike MuskieCrappieBluegillChannel cat

In the box

FlippingFrogsChatterbaits Texas rigNed rigGlide baits CrankbaitsInline spinnersTournaments
~/ideas

AI fishing sunglasses

The combination I'm most excited about: sunglasses that tie into a live sonar / fish-finder feed and use AI to read the water — telling you where the fish actually are, in your field of view, in real time. The server room and the swamp, finally in the same device.

live sonaron-device modelCV overlayyou, catching more fish
[ now playing ] 0x07

The track on loop

◈ SONAR / AUDIO-REACTIVEpress play
the one on loop right now
heavy bass · underground · sped-up

Heavy bass, underground energy

Late-2010s / early-2020s underground SoundCloud sound — aggressive, sped-up, low on traditional vocals, high on low end. Hit play; the sonar reacts to the bass.
On rotation
BronclairPixyPradabagshawty FengKKamiistarKevin Kazi YhapojjFree ShineComethazineChief Keef
heavy basssped-upaggressivelow vocals
~/mission 0x08

Build software. Ship products. Eventually build something huge.

The long game is entrepreneurship: turning technical projects into real products, making things online, and building toward a company at serious scale. I'd rather learn how to turn an experiment into a business than keep every experiment in a folder.

Ship SaaS
Take the projects that work and turn them into products people can actually use.
$1B company
The absurd north star. Aim there, and the smaller wins take care of themselves.
Own presence
Build my own internet footprint and reach out to companies — Launchux, Far Reach, whoever's next.
~/connect 0x09

Find me on the network