1000+
Projects & References
Successful digital solutions across industries — where aesthetics and engineering meet.
Since 2018, blending aesthetics with engineering — building flawless digital experiences and autonomous systems for brands.
Since 2018, blending aesthetics with engineering — building flawless digital experiences and autonomous systems for brands.
No system that doesn't look perfect can have lasting impact. I unite the power of design with the intelligence of software.
I'm Kuzey Yazıcıoğlu, based in Ankara. My path into the digital world began with a passion for visuals and graphic design. After turning a hobby into a professional career and teaching, I pursued software engineering at Çankaya University to bring the worlds I design to life technically. Today, I combine aesthetic vision with deep craft to build products that don't just work — they captivate.
Currently open to corporate collaborations, global freelance projects, and innovative proposals.
Projects & References
Successful digital solutions across industries — where aesthetics and engineering meet.
Since
Continuous production at the intersection of design and software — setting trends, not following them.
Pixel Perfection
Flawless proportions, golden-ratio math, and fluid user experience (UI/UX) in every interface.
Autonomous Systems
Smart architectures with AI integration that accelerate workflows and run on a zero-error principle.

Yoru is a full-screen listening environment built around 21 anime-style scenes and a 13-layer soundscape. Scenes and audio are paired — when you change the track, visuals adapt automatically. After a few seconds of stillness, the interface fades away; nothing stands between you and the atmosphere. Every transition crossfades, every interaction is keyboard-driven; the entire experience is designed to make you forget you're in a browser.
I wanted to build something that delivers a feeling — not productivity or metrics. Most music sites are built around playlists and controls. I wondered what would happen if you removed all of that and designed around atmosphere instead.
PR Sensei connects to GitHub webhooks, captures pull requests, queues them through an async Redis and BullMQ pipeline, and produces structured code reviews using OpenAI and Gemini APIs — all in under 30 seconds. It leaves a summary comment and up to 5 inline comments on changed lines. Reviews are deduplicated; the same line isn't flagged twice, and the system is idempotent per commit SHA. The Next.js dashboard tracks per-repo review metrics, file hotspots, and history backed by a multi-tenant PostgreSQL schema.
Code reviews were a bottleneck on every team I was part of. I wanted something fast and structured to catch obvious issues before the architecture and design conversation — not to replace human review.
LLM Cookbook matches natural-language queries to recipes via FAISS vector search and MiniLM embeddings — when you search for "something warm and spicy," results come from meaning, not keywords. If there's no good match, it generates a recipe from scratch with a local LLM. Supports diet filters, allergen exclusions, and ingredient substitution. A Pandas pipeline converts 10,000+ raw recipe records into model-ready embeddings via a single repeatable script; served through a FastAPI backend.
I wanted to see how far semantic search could go without a full LLM. Recipes are an ideal domain — structured enough to test retrieval quality, complex enough to require real NLP. It started as a class project; I went deep on embeddings, vector search, and when generation vs. retrieval is enough.
Horizon connects to multiple bank accounts via Plaid, pulls real transaction data, and unifies balances, spending breakdowns, transaction history, and fund transfers via Dwolla in a single panel. Authentication is managed server-side with Appwrite; the interface updates in real time as accounts connect or transactions arrive. The focus wasn't just a feature demo — but a full product with proper loading states, responsive tables, Zod form validation, and a consistent design system.
I wanted to build something that feels like a real product, not just a feature. Most portfolio projects stay on the surface — I wanted to wrestle with real bank APIs, real auth flows, real-time data, and the messy details that make an end-to-end app work.

Yoru is a full-screen listening environment built around 21 anime-style scenes and a 13-layer soundscape. Scenes and audio are paired — when you change the track, visuals adapt automatically. After a few seconds of stillness, the interface fades away; nothing stands between you and the atmosphere. Every transition crossfades, every interaction is keyboard-driven; the entire experience is designed to make you forget you're in a browser.
I wanted to build something that delivers a feeling — not productivity or metrics. Most music sites are built around playlists and controls. I wondered what would happen if you removed all of that and designed around atmosphere instead.
PR Sensei connects to GitHub webhooks, captures pull requests, queues them through an async Redis and BullMQ pipeline, and produces structured code reviews using OpenAI and Gemini APIs — all in under 30 seconds. It leaves a summary comment and up to 5 inline comments on changed lines. Reviews are deduplicated; the same line isn't flagged twice, and the system is idempotent per commit SHA. The Next.js dashboard tracks per-repo review metrics, file hotspots, and history backed by a multi-tenant PostgreSQL schema.
Code reviews were a bottleneck on every team I was part of. I wanted something fast and structured to catch obvious issues before the architecture and design conversation — not to replace human review.
LLM Cookbook matches natural-language queries to recipes via FAISS vector search and MiniLM embeddings — when you search for "something warm and spicy," results come from meaning, not keywords. If there's no good match, it generates a recipe from scratch with a local LLM. Supports diet filters, allergen exclusions, and ingredient substitution. A Pandas pipeline converts 10,000+ raw recipe records into model-ready embeddings via a single repeatable script; served through a FastAPI backend.
I wanted to see how far semantic search could go without a full LLM. Recipes are an ideal domain — structured enough to test retrieval quality, complex enough to require real NLP. It started as a class project; I went deep on embeddings, vector search, and when generation vs. retrieval is enough.
Horizon connects to multiple bank accounts via Plaid, pulls real transaction data, and unifies balances, spending breakdowns, transaction history, and fund transfers via Dwolla in a single panel. Authentication is managed server-side with Appwrite; the interface updates in real time as accounts connect or transactions arrive. The focus wasn't just a feature demo — but a full product with proper loading states, responsive tables, Zod form validation, and a consistent design system.
I wanted to build something that feels like a real product, not just a feature. Most portfolio projects stay on the surface — I wanted to wrestle with real bank APIs, real auth flows, real-time data, and the messy details that make an end-to-end app work.
