Profile picture Fatih Karahan

Fatih Karahan

AI Engineer

Oldenburg, Germany

AI Engineer with a deep curiosity for how things work and a strong belief in open source. Focused on building robust AI systems that bridge research and real-world impact.

$ npx connect
Initializing...
Ready to connect.
Waiting for contact request...
interface Developer {
    name: string;
    skills: string[];
    interests: string[];
};

interface TechStack {
    agentAI: string[];
    devOps: string[];
    backend: string[];
    frontend: string[];
    dataEngineer: string[];
};

type ContactType = 'github' | 'youtube' | 'linkedIn' | 'email';

const techStack: TechStack = {
    agentAI: [
        'Pi Coding Agent SDK',
        'OpenCode',
        'Codex',
        'Flue',
        'agent-browser',
        'Pydantic AI'
    ],
    devOps: [
        'Docker',
        'Docker Compose',
        'GitHub Actions',
        'Cloudflare',
        'MinIO',
        'LocalStack'
    ],
    backend: [
        'FastAPI',
        'Node.js',
        'PostgreSQL',
        'DuckDB',
        'SQLAlchemy',
        'Alembic',
        'Airflow',
        'Go'
    ],
    frontend: [
        'TypeScript',
        'Astro',
        'TailwindCSS',
        'React'
    ],
    dataEngineer: [
        'Python',
        'DuckDB',
        'Parquet',
        'Apache Airflow',
        'dlt',
        'Metabase',
        'Grafana',
        'MinIO / S3'
    ]
};

// Developer profile
const me: Developer = {
    name: 'Fatih Karahan',
    skills: [
        ...techStack.agentAI,
        ...techStack.devOps,
        ...techStack.backend,
        ...techStack.frontend,
        ...techStack.dataEngineer
    ],
    interests: [
        'AI Agents & Orchestration',
        'Coding Agent Infrastructure',
        'Open Source',
        'Browser Automation',
        'Developer Tooling',
        'Data Engineering',
        'Technical Content Creation'
    ]
};

// Contact function
const contact = (type: ContactType): string => {
    switch (type) {
        case 'github':
            return 'https://github.com/Fatih0234';
        case 'youtube':
            return 'https://youtube.com/@StrataAcademy';
        case 'linkedIn':
            return 'https://www.linkedin.com/in/fatih-karahan-717931193/';
        case 'email':
            return 'sekanti02@gmail.com';
        default:
            return 'Send fax.';
    }
};

Maestro

A minimal orchestrator for AI coding agents. Polls a local board, creates git-worktree workspaces, dispatches agents through a plan-execute-verify pipeline, and monitors progress via a Bubble Tea TUI. Built with Go and the Charm stack.

StudIP Browser Agent

A local browser agent for Stud.IP — natural-language notice board management and degree programme assistance. Uses Flue for agent orchestration and agent-browser for browser control. Features saved auth state, typed tools, and structured Valibot outputs.

CHM — Core Inventory & Run Health

A monitoring backend for data engineering teams to track client pipelines, ingest run events idempotently, and provide dashboard-ready operational health views. Built with FastAPI, SQLAlchemy, PostgreSQL, and Grafana.

GitPulse

A GitHub issue backlog analytics pipeline answering whether the open issue backlog is growing or shrinking. Bronze → Silver → Gold data architecture using DuckDB, Parquet, MinIO, Apache Airflow 3.0, and Metabase.

Research Assistant

Carl von Ossietzky University of Oldenburg — VLBA

Dec 2024 — Apr 2026

Oldenburg, Germany

  • Co-authored a peer-reviewed paper for Springer on AI-supported cyclist detection using YOLO11n — gained end-to-end experience in academic research and scientific publishing
  • Self-initiated the curation of a custom 5,000+ image dataset on Roboflow — introducing a novel three-class cyclist split (back, front, side) to address model blind spots on side-view detections, an insight missing from existing benchmarks
  • Integrated a pre-trained YOLO model as an ML backend into Label Studio via HTTP API, transforming the annotation pipeline from manual frame-by-frame labeling to verification-only — reducing annotation time by ~80–90%
  • Trained and fine-tuned YOLO11n models on the curated dataset, achieving 93.4% mAP50 with real-time inference (2.2ms/image on T4 GPU)
  • Designed and built an ETL data pipeline managing heterogeneous multi-sensor data (video, LiDAR, radar, ultrasonic, thermal, GPS) through MinIO object storage, DuckDB, and PostgreSQL
  • Developed an interactive Streamlit dashboard for sensor data quality analysis and visualization
  • Contributed to the BikeDetect project (€200K, funded by German Federal Ministry mFUND), including real-traffic field tests in Osnabrück
Python YOLO11n Label Studio Roboflow Computer Vision Deep Learning MinIO PostgreSQL DuckDB Streamlit FastAPI Docker NGINX Git
Built with Astro by Fatih Karahan