I’m Ismail, an AI engineer.

I train models, investigate what they learn, and build applications. I also create educational content about deep learning.

~/projects

ismail@lab:~$ ls -la projects/

AI Interpretability Journey
I train small Transformers on minimum and median tasks, then inspect attention and weight patterns to understand what they learn.
Simpsons Character X-Ray ↗
A team project exploring multiple model approaches to recognize Simpsons characters in images.
Venture · AI-assisted legal workflow product
Dekasis — AI-assisted workflow for law firms
My venture brings document intake, classification, data extraction, and verification for motor vehicle diminished-value cases into one workspace. AI prepares a first draft; critical details are checked against the source documents.

interpretability/find-min — notebook output

AI_INTERPRETABILITY_JOURNEY / FIND-MIN

AI Interpretability Journey

I train small Transformers on minimum and median tasks, then inspect attention and weight patterns to understand what they learn.

Model selection, training, evaluation, and investigating what the models learned.

OV CIRCUIT · LAYER 0 / HEAD 0Sequence length

seq_len=2 · max_num=50Train 60,000 · validation 20,000

Images come from notebook outputs. Each plot uses its own color scale, so brightness is not a direct numeric comparison across models.

learning.log

Start by building a feel for why an idea works, then work through the mathematics and implementation. This is how I learn and teach.

▶ Yeni Nesil Karar Modelleri: Jev, Clef ve Laya

personal/

Music I love, lyrics I translate, and things I want to write about.

Turkish translations of songs by Orelsan. Each entry links to the Notion page where it was published.