# Hasan Basbunar

> Artificial intelligence engineer: Forward Deployed AI Engineer at Liberty International (Luxembourg). Specialises in LLMs, VLMs, the Model Context Protocol, autonomous AI agents, and LLM fine-tuning. Bilingual French/English portfolio at [hasanbasbunar.fr](https://hasanbasbunar.fr).

Published: 2026-07-07 · Last updated: 2026-07-21

## Who

- Name: Hasan Basbunar
- Current role: Forward Deployed AI Engineer at Liberty International, Luxembourg (since May 2026). Deployed inside the group's businesses (pharma, hospitality, media, equestrian) to automate workflows and ship AI agents to production.
- Based in: Luxembourg and Grand Est (Lorraine), France; cross-border profile in the Greater Region (Grande Région: Luxembourg / Lorraine).
- Role in French: ingénieur en intelligence artificielle (ingénieur IA), expert LLM, VLM et MCP.
- Languages: French (native), Turkish (native), English (professional), German (A2/B1).
- Contact: basbunarhasan@gmail.com
- Site: [hasanbasbunar.fr/fr](https://hasanbasbunar.fr/fr) (French) · [hasanbasbunar.fr/en](https://hasanbasbunar.fr/en) (English)

## Experience

- Liberty International (Luxembourg), Forward Deployed AI Engineer, 2026-present: AI agents, MCP, LLMs/VLMs, RAG, tool use, automation.
- Cleva (Paris), AI Engineer & Tech Lead (apprenticeship), 2024-2026: autonomous AI agents, Model Context Protocol, VLM-based document extraction, async Python in production.
- Foyer Assurances (Luxembourg), Data Scientist intern, 2024: VLMs for document processing. Runner-up, SD-SIF Best Data Science Internship Prize 2025.
- LEM3 / CNRS (Metz), Machine Learning intern, 2022: CNN / U-Net for material image reconstruction.

## Selected work

- [Lodos-24B](https://huggingface.co/hasanbasbunar/Lodos-24B-Instruct-2510): full fine-tuning of a 24B-parameter multimodal LLM (Mistral Small 3.2) on a 20M+ token Turkish corpus.
- Linux kernel: four patch series submitted upstream, two merged to mainline (page_pool and others), reviewed and tested by maintainers.
- [Talkink](https://talkink.app): 100% on-device push-to-talk dictation for macOS (Apple MLX), three ASR engines, distributed via Homebrew.
- Constat amiable VLM: QLoRA fine-tune of Qwen3-VL 8B for structured data extraction from insurance claim forms.
- Homelab: DGX Spark GB10 128GB (Asus Ascent GX10) + MacBook M4 + RTX 4070 Ti Super, exo cluster for home-grown distributed inference.

## Skills

LLMs, VLMs, Model Context Protocol (MCP), autonomous agents, fine-tuning (full + QLoRA), RAG, Python (async, typing), PyTorch, Transformers, vLLM, llama.cpp, quantization (GGUF, NVFP4), Unsloth/TRL, Gradio, FastAPI, Docker.

## Links

- [LinkedIn](https://www.linkedin.com/in/hasanbasbunar/)
- [GitHub](https://github.com/hasso5703)
- [Hugging Face](https://huggingface.co/hasanbasbunar)