I build AI systems for the energy sector. İstanbul.
Most of my work is forecasting and document understanding.
Forecasting — ensemble models predicting generation for wind and solar plants, where being confidently wrong costs real money and the weather has no respect for your validation split.
Document understanding — LLM-based extraction pipelines that read unstructured paperwork and pull out the fields a person would otherwise retype by hand.
The rest is plumbing of the useful kind: a pricing engine, automatic classification and routing for incoming requests, and agent systems built with LangGraph and n8n.
I also hold a Product Owner role on one of our teams, which has mostly taught me how much of engineering is deciding what not to build.
About two and a half years in, heading deliberately toward AI and agentic engineering.
Python · LangGraph · LangChain · n8n · PyTorch · scikit-learn · FastAPI · Docker · PostgreSQL · Java / Spring Boot · TypeScript / Angular
On my own projects: Swift / SwiftUI · TypeScript
The professional work lives in private repositories. What's public is what I build on my own time, under LJASP, and an archive of university coursework.
Building now
- Cumba — an open-source (MIT) macOS app that
turns the MacBook notch into a panel: a mini player for whatever is playing, Claude Code
and Codex progress and approvals, widgets in tabs, and themes. Written in Swift, sends no
telemetry. cumbanotch.com ·
brew install --cask ljasp-studio/cumba/cumba - LJASP — a deterministic football management simulation in the browser: the engine plays out every decision the same way each time. The code is private; the daily puzzle below comes from it.
Archive, 2022-2023
Eight separate assignment repositories are now one — the work moved rather than went away — and some of it contains instructive failures I'd rather document than hide.
- ml-coursework — eight ML assignments collected into one repository: GANs, GRUs, autoencoders, dimensionality reduction. The READMEs say which models didn't work and why, which is the only part still worth reading.
- infomanager-backend · infomanager-frontend — a Spring Boot + Angular CRUD application from a 2022 internship.
- DropboxCryptor — AES-encrypted Dropbox client with post-download integrity checking. 2022.
From LJASP, a deterministic football management simulation I build. A new decision every day.
Black XI 1 – 3 Yellow XI · 68'
Into the final stretch with chances at a premium and the next change likely to tilt momentum.
What's your call?
Reveal
Call "Introduce a fresh striker for penalty-box threat" and the engine boosts momentum (+12), sharpness (+11) and morale (+10) at no added risk. A new centre-forward takes the stage, hugging the last line and attacking the six-yard channel.
Seeded by date. Last run: 2026-11-12.
The snake is playable: steer it yourself.