Niklas Muennighoff
AI Research at Cursor
Palo Alto, California, United States
Summary
🤩
Rockstar🎓
Top SchoolNiklas Muennighoff is an AI researcher based in Palo Alto with six years of hands-on experience building and evaluating large language models and ML infrastructure. He has contributed to high-profile open-source projects like EleutherAI’s lm-evaluation-harness and the MTEB embeddings benchmark, adding ethical-evaluation tasks and fixing core scoring logic to improve model assessment. His industry research roles span Hugging Face, Aleph Alpha, AI2, Meta, and currently Cursor, with multiple peer-reviewed and preprint publications linked to those appointments. Niklas blends research rigor with engineering practice—implementing data pipelines, efficient inference hooks (e.g., VLLM), and evaluation tooling for code and text generation. He also brings an uncommon background in voice acting and finance education, reflecting strong communication skills and interdisciplinary perspective. This combination makes him adept at translating ML research into robust evaluation frameworks and reproducible engineering.
6 years of coding experience
Bachelor, Business/Finance, Bachelor, Business/Finance at Peking UniversityPre-college Program, Astrophysics, Pre-college Program, Astrophysics at Harvard UniversityPhD, Computer Science, PhD, Computer Science at Stanford University
English, Japanese, Chinese
Github Skills (26)
sentence-transformers10
embed10
python10
word-embeddings10
machine-learning10
text-classification10
semantic-search10
trainings10
natural-language-processing10
evaluation-framework10
nlp10
bleu10
embedding10
modeling10
transformers9
Programming languages (10)
Github contributions (5)
simplescaling/s1
Feb 2025 - Apr 2025
s1: Simple test-time scaling
Role in this project:
ML Engineer
ML EngineerContributions:14 reviews, 10 PRs, 44 pushes in 1 month
Contributions summary:Niklas's commits primarily involve modifications to data loading, preprocessing, and model training scripts, specifically `data/collect_data.py` and `train/sft.py`. These changes include updates to dataset paths and loading mechanisms for open-source math datasets, indicating a focus on preparing data for model training. Further modifications to training configurations suggest involvement in model experimentation and refinement. The integration with VLLM, as seen in `eval/generate.py`, suggests a focus on efficient inference.
embeddings-benchmark/mteb
Jul 2022 - Jan 2023
MTEB: Massive Text Embedding Benchmark
Role in this project:Back-end Developer
Back-end DeveloperContributions:7 releases, 255 reviews, 153 commits in 6 months
Contributions summary:Niklas primarily focused on fixing issues related to the main score calculation in a multilingual text embedding benchmark. They addressed warnings and made adjustments to the `AbsTaskClassification.py` file, ensuring the correct handling of main scores. Furthermore, the user updated various task configurations in multiple files to correctly set main scores and fix task splits for accurate evaluation. The user also refactored the summarization evaluator and adjusted the code to skip samples with no variance.
bertsimilarity-searchbenchmarkretrievalbug-reporting
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.