Hi, I’m Delip Rao

I am an AI researcher, author, and founder. I have worked on machine learning and natural language processing for two decades, across three generations of the field: statistical NLP during my graduate research at Johns Hopkins, deep learning in industry, and today’s large language models. I am currently at the University of Pennsylvania.

Now

My main research is non-verifiable problems, especially in the AI for Science context, and the role of rubrics in agentic evaluation and reward modeling. I also work on other topics like building large-scale datasets, mitigating hallucinations, verifying scientific claims, and building agentic workflows for science. See Research and Publications.

Industry

I was a research scientist at Amazon, working on natural language understanding for what became Alexa. As Vice President of Research at AI Foundation, I led teams generating and detecting synthetic media across speech, language, and vision. I was Head of Machine Learning at Polywork.

Companies I built

In 2014, I founded Joostware AI Research, one of the first companies to bring deep learning research consulting to Fortune 500 companies. I went on to build R7 Speech Sciences, which extracted insights from spoken conversations, and MangoMind AI, which applied retrieval-augmented generation to healthcare. I also co-founded FakeNewsChallenge.org, a non-profit that organized AI researchers against misinformation.

Writing and teaching

I wrote Natural Language Processing with PyTorch (O’Reilly), used by practitioners in industry and students in academia worldwide, and translated into German and Chinese. I write a newsletter on AI research and strategy with a few thousand subscribers.

If you are building something ambitious with AI, I would love to hear from you.