Alperen Uğur Erden
physics undergrad at Bilkent · Ankara, Turkey (UTC+3) · remote
Flagship · Systems that run themselves live · family-run

Greenhouse

IoT + computer vision

My dad grows dragon fruit in a greenhouse near Bursa. I put the whole thing into one app the family actually uses: five cameras, a temperature and humidity sensor, one-tap watering, and a frost alarm that warns everyone's phone before a cold night can kill the crop. An ESP32 reads the sensors, a small AI model watches the cameras, and Cloudflare ties it all together.

family-only app · happy to give a live walkthrough

I build things that run in the real world: AI agents, computer vision, and full-stack apps. Mostly solo, mostly deployed.

Most of what I make is live and running in production. A greenhouse my family actually uses. A notes platform thousands of students rely on. Agent systems that check their own work. I taught myself the stack, and I ship fast.

Alperen Uğur Erden, portrait at night in Göreme
Göreme, at night
Availability
Open to remote contract & summer-2026 internship
Get in touch alperenugur04@gmail.com → Main site sirschrodinger.com ↗ GitHub github.com/sirSchrodinger ↗

Systems that run themselves

built once, still running without me greenhouse featured above ↑
AI agentsbuilt & run

papers-net & karpathy

Two AI systems I built and ran for months. papers-net writes research papers, then checks its own work against four different AI models to catch mistakes before they slip through. karpathy sets its own research goals and verifies what it finds. I have paused both, but the paper series they produced is still online.

papers.sirschrodinger.com ↗
quanthonest negative result

Trading stack

A crypto trading bot with a risk layer that bets more when the model is sure and less when it isn't. I test every signal hard and throw out whatever doesn't hold up. When the edge came out flat, I wrote it up honestly as a paper with the negative result instead of burying it.

private dashboard

Products with real users

live sites people actually use
platformlive2,500+ notes

defter · thelecturenotes.com

A free, ad-free library of lecture notes for my university. Over 2,500 documents indexed, real students using it every day, and a pipeline that turns messy uploads into clean notes. No paywall, and cheap to keep running.

thelecturenotes.com ↗
marketplacelive

yurttanayriliyorum.com

A marketplace where Turkish students hand off their things when they move out of the dorms. Shipped solo, with student-email sign-in, a separate site per campus, automatic screening of inappropriate photos, and the data-protection paperwork most side projects skip.

yurttanayriliyorum.com ↗
legal techlive

davamcepte

Built with a lawyer friend: a place where someone can follow their own court case without calling the office. The case history in plain Turkish, a countdown to the next hearing, and a glossary that translates the legal jargon.

davamcepte.sirschrodinger.com ↗
personal sitelive

sirschrodinger.com

My personal site, built like a tiny operating system. A 3D room you can walk into, a live music widget, an admin panel with real analytics behind Google sign-in, and a chatbot trained on my own messages. Cloudflare Workers, D1, and Three.js.

sirschrodinger.com ↗

Applied ML / computer vision

models put to work on real data
sports videoin progress

Halısaha-stats

Point a phone at a five-a-side football match and it works out the stats on its own: who is who, how far each player ran, a heatmap, and a bird's-eye replay. It finds and tracks the players, corrects the fisheye lens, and flattens the pitch to a top-down view. I measure how accurate it is against real ground truth instead of trusting a nice-looking demo.

building toward a product
medical MLresearch paper

Sclerotic radiomics

With a radiologist and a physician, a model that tells dangerous bone tumors apart from harmless bone spots on a CT scan. 180 lesions, careful patient-grouped cross-validation, around 0.985 AUC. I caught a subtle data leak in my own pipeline with a cross-model review and fixed it before the number could lie.

paper
searchprivate

Immich OCR

I ran text recognition over 59,000 family photos so every word inside them became searchable, then used that to pull specific legal documents out of a decade of mess automatically. Self-hosted, with full-text search that handles Turkish accents. (Private data, happy to walk through how it works.)

private
voice AIruns locally

Berin

A voice tutor for practicing technical interviews. It reads the question out loud, transcribes my spoken answer with local speech recognition, grades it on five points with an AI, then brings my weak spots back on a spaced-repetition schedule. Runs on my own machine.

runs locally
Also

bsc demo & physics papers

A 3Blue1Brown-style interactive walkthrough of one of my cosmology papers (React-Three-Fiber and KaTeX), plus math-physics drafts I write with my own LaTeX tooling.

sirschrodinger.com/ai_projects/bsc/ ↗
About

I study physics at Bilkent and taught myself everything on the software side, mostly by building things at hours I should have been asleep.

What I keep coming back to is software that touches the real world and then keeps running on its own: a greenhouse that texts the family before a frost, agents that set their own goals and check each other's work, infrastructure that costs almost nothing to keep alive. I work solo and fast. I'd rather have a rough thing running tonight than a perfect plan for next week.

I also care about the parts nobody screenshots: sign-in, moderation, data protection, not falling over at 3am. That is the difference between a demo and something people actually use.

Stack