Sam Shleifer

2022: Research Engineer at startup training large language models 2021: Research Engineer at FAIR training large language models + writing papers 2020: Research Engineer at HuggingFace working on distilling huge transformer models and improving sequence2sequence support (tasks like summarization, translation, dialogue) for the most popular open source NLP library in the whole wide world https://github.com/huggingface/transformers 2019: Moved to SF. Worked on suggesting responses to doctors in doctor/patient chats at Curai https://arxiv.org/abs/1910.03476 and some research on Data Augmentation Tricks for Text Classification https://bit.ly/2CGvXYl at Stanford. 2015-2018: My industry experience is evenly split between ML on timeseries, image and text data. At Kensho, where I spent the last 4 years, I built and managed a 5 engineer team, but got to still spend most of my time running experiments and coding. Our product predicted demand for securities using proprietary transaction data and historical market data. After Kensho got bought by S&P Global, I started working on entity linking from databases and free text, using ML. I’ve also worked on computer vision on medical scans. At Merantix, a startup in Berlin, we used faster-rcnn to find tumors in breast scans. On Kaggle, I worked on finding lung occlusions in pneumonia patients using a combination of retinanet and mask-rcnn. I also got a silver medal in a Kaggle competition that was more traditional Natural Language Understanding: finding duplicate Quora questions using bidirectional LSTMs. Before Kensho, I graduated from Yale ('15), where my Economics thesis, which analyzed Peer to Peer lending, won the prize for the best undergraduate finance thesis. I've had internships with an education non-profit, two pro sports teams, and a hedge fund in Brazil. Undergraduate Thesis (November '14- April '15): github.com/sshleifer/ProsperThesis