Raj Movva

I am wrapping up my PhD at Berkeley AI, where I am advised by Emma Pierson and supported by the NSF GRFP.

My goal is to build AI that is (i) safe and aligned and (ii) accelerates science, especially in medicine and biology. To these ends, my PhD focused on two topics: LLM post-training, and interpretability-guided scientific discovery. Some work I'm proud of:

1. HypotheSAEs: we train SAEs on embeddings to generate scientific hypotheses from text datasets.
2. What's In My Human Feedback: we automatically learn latent preferences encoded in human feedback, enabling more robust and controllable RLHF.
3. Work in progress: we interpret medical foundation models to learn biomarkers that doctors miss.

Previously: I did my undergrad in CS at MIT, and worked with Catherine D'Ignazio, Michael Carbin, and Anshul Kundaje. I've interned at Microsoft Research, Apple ML, NVIDIA, and Genesis Therapeutics. I'm deeply passionate about biology, though I haven't worked on it as much lately.

Links

• Email: rmovva@berkeley.edu
• Scholar: Rajiv Movva
• GitHub: @rmovva
• Twitter: @rajivmovva
• CV: pdf
• Collaborators and mentors I've been lucky to work with