Doğukan Ürker
full-stack engineer at sensity.ai — building detection systems for deepfakes & ai-generated media across backend and frontend. off the clock i benchmark local llms, build open-source evaluation tooling, and publish reproducible model configs, and findings. i also co-lead engineering for gdg on campus yaşar and maintain open-source projects with 59k+ combined downloads. coding since 12, still chasing the same feeling.
experience
Sensity AI
full-stack engineer
aug 2025 – present
remote (eu)
- ship full-stack features across FastAPI microservices and a React & Vue frontend
- help migrate legacy Django services onto the modern FastAPI stack
- build internal review platforms, developer-facing tooling, and integration tests
- review pull requests across both frontend and backend
Sensity AI
backend intern
mar 2025 – aug 2025
remote (eu)
- five-month backend internship focused on FastAPI microservices; converted to full-time engineering
GDG on Campus Yaşar University
software team, core member
sep 2024 – present
izmir
- co-lead a six-engineer software team; own architecture and task distribution alongside the team lead
- designed the monorepo from scratch: 4 FastAPI microservices (user, form, mail, event) and 3 React frontends
- defined team conventions: tooling (uv, bun, Ruff, Biome), directory structure, git workflow, conventional commits, squash-merge policy
- break initiatives into issues, distribute tasks, review pull requests, mentor contributors
- stack: Python 3.14, FastAPI, Motor (MongoDB), React 19, TypeScript, Vite, TailwindCSS
selected projects
a handful from 105+ things i've built since i was 12
Tamga
open-source python logging library
colorful tailwind-inspired console output, file/json logging with rotation, sqlite and mongodb integration, and email notifications for critical logs — used across personal projects and in production at gdg yaşar.
FlaskBlog
full-stack blogging platform
self-hostable flask blog with authentication, post management, and a responsive ui; adopted for real-world deployments.
BenchKit
evaluation harness for local llms
terminal ui and headless runner across 15 suites, including humaneval+, mbpp+, gpqa diamond, ifeval, and ruler — with degeneration diagnostics, per-item run diffing, decontamination tooling, and inference profiling. drives every benchmark result i publish and works with any openai-compatible server.
DogiZed
theme for the zed editor
minimalist dark/light dual theme with pure black and white backgrounds and vibrant syntax.
Kira
self-hosted personal assistant
telegram frontend with a local 14b llm served via llama-swap on my homeserver — designed around a "dumb llm, smart tooling" architecture: the model only classifies intent while every action, api call, and scheduled task runs in python.
skills
- languages
- python, typescript, javascript, html, css
- backend
- fastapi, flask, django, motor, pydantic, async/await patterns
- frontend
- react 19, vue, typescript, vite, tailwindcss, shadcn
- ai / ml
- llama.cpp / llama-swap, gguf quantization, qlora fine-tuning (unsloth), llm evaluation & benchmark design, contamination analysis, pytorch
- infrastructure
- docker, tailscale, vast.ai, makefile, systemd, debian
- databases
- mongodb, sqlite
- tooling
- git, uv, bun, ruff, biome, zed, claude code
education
Yaşar University
b.sc. software engineering
languages
turkish (native) · english (c1)