yusuf@system — zsh
$

AI ENGINEER◇MULTI-AGENT SYSTEMS◇COMPUTER VISION◇NLP PIPELINES
[01]OBSS AI INTERN
[02]TÜBİTAK RESEARCHER
SYSTEM ONLINE
SCROLL TO EXECUTE ⌁
yusuf@system — identity
01 / IDENTITYLAYER ALIGNMENT
NAMEYusuf Çınar
ROLEAI Engineer
DOMAINMulti-Agent · Vision · NLP
AFFILIATIONOBSS · TÜBİTAK
/ / Systems that plan, perceive and act. Research-grade computer vision, production NLP pipelines and agentic development workflows.[3 ACTIVE DOMAINS]
yusuf@system — experience
02 / EXPERIENCE
3 PROCESSES

OBSS

2025 —/PROCESS RUNNING
ROLE
AI Intern
INPUT
AI-Native Development
STACK
Claude / Cursor / Codex
METHOD
ATDD / Test-first / Red-team
EXECUTION TRACE
01spec → decomposition → acceptance criteria
02agent build loop · in-editor context
03red-team pass · adversarial review
04test-first verification · regression sweep

MAVİ LOJİSTİK

2024 — 2025/PROCESS COMPLETE
ROLE
AI Developer
INPUT
WhatsApp operations channel
STACK
NLP pipeline / LLM routing / Automation
OUTPUT
Autonomous message triage + response
0
MESSAGES / DAY
0%
AUTOMATION
0
RESPONSE LATENCY

TÜBİTAK

2024/RESEARCH ARCHIVED
ROLE
Researcher
INPUT
20,000+ Sentinel-2 scenes
MODEL
U-Net → DeepLabV3+
TASK
Semantic segmentation / flood mapping
SCENE / SENTINEL-2
MASK / PREDICTION
MODEL COMPARISON
U-NetIoU 62
DeepLabV3+IoU 74
U-Net → DeepLabV3+  ·  IoU +12%  ·  GEN +10%
yusuf@system — projects
03 / PROJECTSHORIZONTAL TRANSPORT △
PROJECT 01SATELLITE

flood-detection

Sentinel-2 flood segmentation over multi-region scenes.

[PyTorch][DeepLabV3+][Rasterio][OpenCV]

/ / U-Net baseline replaced by DeepLabV3+ with ASPP; IoU +12% on held-out regions.

<> OPEN ON GITHUBVIEW REPO
PROJECT 02VISION LANGUAGE

ViLT

Vision-and-language transformer experiments without convolutions.

[Transformers][PyTorch][CLIP][VQA]
FUSION

/ / Patch-embedded image tokens fused with text tokens in a single transformer stack.

<> OPEN ON GITHUBVIEW REPO
PROJECT 03OS

windowsphereAI

An agentic desktop-style environment for AI operations.

[React][TypeScript][LLM Agents][FastAPI]
/system
├── agents/
│ ├── planner.ts
│ └── executor.ts
├── windows/
└── kernel.ts

/ / File-tree addressable agents; each window is a process with its own tool scope.

<> OPEN ON GITHUBVIEW REPO
PROJECT 04DATA

ml-lab

Classical ML pipelines, benchmarking and feature analysis.

[scikit-learn][NumPy][Pandas][Matplotlib]

/ / Reproducible experiment runners with cross-validated metric reporting.

<> OPEN ON GITHUBVIEW REPO
yusuf@system — stack
04 / TECH STACKDEPENDENCY GRAPH
ROOT RUNTIME
Python
[4 MODULES · 9 CAPABILITIES]
PyTorch
├──U-Net
├──DeepLabV3+
└──ViLT
OpenCV
├──Preprocessing
└──Masking
Transformers
├──NLP Pipelines
└──LLM Routing
FastAPI
├──Serving
└──Agents API
yusuf@system — pipeline
05 / AI-NATIVE PIPELINEPLAN → BUILD → VERIFY
/ / SPECIFICATION BEFORE CODE. VERIFICATION AFTER.
yusuf@system — github
06 / GITHUB
DEVELOPER DASHBOARD
04
PUBLIC REPOS
46
TOTAL STARS
Python
PRIMARY LANG
REPOSITORYACTIVITY
flood-detection[RESEARCH]RECENTViLT[EXPERIMENT]2 MONTHSwindowsphereAI[SYSTEM]ACTIVEml-lab[NOTEBOOKS]6 MONTHS

/ / Full source, experiments and research code live on the public profile.

<> GITHUB.COM/YUSUFCINARCI