Noé Flandre NoeFlandre

Noé Flandre

AI Research Engineer · Toulouse, France · vibe-learning
Daily meal: curating datasets & training models 🍜🤖

LLM data: harvesting → filtering → labeling → training (and pushing it on HF 🤗).
Multimodal: VLM fine-tuning, evaluation, and “does it actually work in the wild?” benchmarks.
Maximizing my learning curve: One chaï and one video of king Karpathy and I got my weekend covered.

🏛️ FineWeb-Legal : legal-domain extraction pipeline + classifier-trained filtering
→ repo: https://github.com/NoeFlandre/fineweb-legal
→ datasets/models: https://huggingface.co/NoeFlandre
🧠 GPT-2 From Scratch : learning by building (DDP, training, eval, pain & joy thanks to Karpathy)
→ https://github.com/NoeFlandre/GPT2-From-Scratch
🧠 Transformer From Scratch : learning by building (again), following Umar Jamil tutorial
→ https://github.com/NoeFlandre/transformer-from-scratch
🗺️ mini-geo-parse : very simple and tiny geoparsing pipeline (LLMs → locations → coordinates)
→ https://github.com/NoeFlandre/mini-geo-parse
🖼️ nanoclip : CLIP from scratch + training experiments
→ https://github.com/NoeFlandre/nanoclip
🧨 JailBreak-DeepSeek : jailbreak robustness evaluation playground
→ https://github.com/NoeFlandre/JailBreak-DeepSeek

Promoting empathy in decision-making by turning agent-based models into stories using large-language models (Journal of Simulation, 2025) : https://www.researchgate.net/publication/395240074_Promoting_empathy_in_decision-making_by_turning_agent-based_models_into_stories_using_large-language_models
Can Large Language Models Learn Conceptual Modeling by Looking at Slide Decks and Pass Graduate Examinations? An Empirical Study (EmpER @ ER, 2024) : https://link.springer.com/chapter/10.1007/978-3-031-75599-6_15

One place, two views: the core idea behind GeoReasoner : https://noeflandre.com/posts/georeasoner
Building FineWeb-Legal: A 10B Token Pilot : https://noeflandre.com/posts/fineweb-legal
Turning agent-based models into empathetic stories (without getting poetic) : https://noeflandre.com/posts/promoting_empathy
Can LLMs learn conceptual modeling from slide decks? : https://noeflandre.com/posts/llms-conceptual-modeling

If you’re building LLMs, data pipelines, VLM fine-tuning, or anything that smells like “AI research”, I’m in.