Anuroop Sriram

Anuroop Sriram

Founding AI Research Scientist, Prometheus

I build AI foundation models, with a focus on using them to accelerate science and engineering. I am a founding AI Research Scientist at Prometheus, where we are building the Artificial General Engineer: AI systems that change how the physical world is designed and engineered.

Before Prometheus, I was a technical lead in the FAIR Chemistry team at Meta, where I worked on universal machine-learning interatomic potentials (UMA), generative models for materials (FlowLLM, FlowMM), and some of the largest open datasets in the field, including Open Catalyst and Open DAC. I also created FastCSP, which cuts crystal structure prediction from weeks to hours. I helped build the fairchem library and invented Graph Parallelism, the distributed-training method behind today's largest interatomic potentials, used to train models at billion-parameter scale.

I also led fastMRI, a collaboration between Meta and NYU Langone to accelerate MRI using deep learning. fastMRI delivers up to 4× faster scans, and methods based on it have been deployed in clinical MRI systems worldwide.

Earlier, I built and led the speech team at Meta FAIR, where we trained some of the first billion-parameter speech models and built highly efficient ASR systems that were deployed to billions of users. Before that, at Baidu, I co-created Deep Speech 2, the first end-to-end model to achieve human-level speech recognition in both English and Mandarin.

I hold a Master's in Language Technologies from Carnegie Mellon University.

Selected Publications

  • UMA: A Family of Universal Models for Atoms
  • UMA: A Family of Universal Models for Atoms
    Brandon M. Wood, Misko Dzamba, Xiang Fu, et al.
    NeurIPS 2025
    ARXIV CODE MODELS
  • FlowMM: Generating Materials with Riemannian Flow Matching
  • FlowMM: Generating Materials with Riemannian Flow Matching
    Benjamin Kurt Miller, Ricky T. Q. Chen, Anuroop Sriram, et al.
    International Conference on Machine Learning (ICML) 2024
    ARXIV CODE
  • Towards Training Billion Parameter Graph Neural Networks for Atomic Simulations
  • Towards Training Billion Parameter Graph Neural Networks for Atomic Simulations
    Anuroop Sriram, Abhishek Das, Brandon M Wood, et al.
    International Conference on Learning Representations (ICLR) 2022
    ARXIV CODE
  • Open catalyst 2020 (OC20) dataset and community challenges
  • Open catalyst 2020 (OC20) dataset and community challenges
    Lowik Chanussot, Abhishek Das, Siddharth Goyal, et al.
    ACS Catalysis
    ARXIV CODE DATA
  • End-to-end variational networks for accelerated MRI reconstruction
  • End-to-end variational networks for accelerated MRI reconstruction
    Anuroop Sriram, Jure Zbontar, Tullie Murrell, et al.
    International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2020
    ARXIV CODE
  • fastMRI: A publicly available raw k-space and DICOM dataset of knee images for accelerated MR image reconstruction using machine learning
  • fastMRI: A publicly available raw k-space and DICOM dataset of knee images for accelerated MR image reconstruction using machine learning
    Florian Knoll, Jure Zbontar, Anuroop Sriram, et al.
    Radiology: Artificial Intelligence
    ARXIV CODE DATA
  • Cold Fusion: Training Seq2Seq Models Together with Language Models
  • Cold Fusion: Training Seq2Seq Models Together with Language Models
    Anuroop Sriram, Heewoo Jun, Sanjeev Satheesh, et al.
    Interspeech 2018
    ARXIV
  • Deep speech 2: End-to-end speech recognition in english and mandarin
  • Deep speech 2: End-to-end speech recognition in english and mandarin
    Dario Amodei, Sundaram Ananthanarayanan, Rishita Anubhai, et al.
    International Conference on Machine Learning (ICML) 2016
    PAPER