Looking for a job

Semih Orhan

@semihorhan

Machine Learning Engineer building LangGraph assistants for automated data analysis and modular RAG pipelines.

I built a domain-agnostic agentic AI data assistant with LangGraph and Google Gemini that turns natural-language questions into automated analysis of relational JSON databases. It discovers schemas at runtime, generates and safely runs Python/Pandas code, and can recover from errors through a retry loop.

I also built a modular RAG pipeline with LangChain, FAISS, SentenceTransformers embeddings, and the OpenAI API. I served the systems through FastAPI and containerized them with Docker.

At TU/e, I developed a monocular vision system for 3D human pose estimation, using model-reduction techniques in PyTorch to double FPS while keeping accuracy. Earlier, I worked on visual localization and panoramic-image semantic segmentation, including releasing a panoramic dataset and open-sourcing a model.

Experience

Work history, roles, and key accomplishments

Current

Machine Learning Engineer

Current

Freelance

Mar 2025 - Present (1 year 7 months)

Built a domain-agnostic agentic AI system with LangGraph and Google Gemini for automated data analysis, and a modular RAG pipeline with LangChain, FAISS, and OpenAI API.

LangGraph Lang Chain RAG Gemini OpenAI API Faiss Sentence Transformers fastAPI Pydantic Docker MkDocs Python Pandas

Machine Learning Engineer & Computer Vision Engineer

TU/e

Oct 2022 - Oct 2024 (2 years)

Developed a YOLO-based monocular vision system for 3D human pose estimation, doubling FPS while maintaining accuracy, and supervised student projects on pedestrian intention prediction and human-robot interaction.

YOLO PyTorch Computer Vision 3D Pose Estimation Model Reduction RNN GRU graph neural networks

PhD Studies on Machine Learning and Computer Vision

IYTE

Jan 2018 - Aug 2022 (4 years 7 months)

Developed a semantic segmentation model for panoramic images with a 4% mIoU improvement, released an outdoor panoramic dataset, and served as a research and teaching assistant.

Semantic Segmentation Panoramic Imaging Equirectangular Projection PyTorch Teaching Research

Machine Learning Engineer & Computer Vision Engineer

University of Zaragoza

Sep 2020 - Sep 2021 (1 year)

Combined self-supervised contrastive semantic features with RGB features in a visual localization pipeline, improving accuracy by 2% and robustness to seasonal changes.

Self Supervised Learning Contrastive Learning Image Based Localization Computer Vision