Foundation Model Virtualization

An operating-system-inspired layer that virtualizes foundation models for enterprise and scientific workloads.

Serving and managing a growing zoo of foundation models at enterprise scale is hard: models differ in cost, capability, and reliability, and workloads change over time. This line of work rethinks foundation model serving through the lens of operating-system principles — introducing a self-evolving OS layer that virtualizes foundation models, dynamically routes requests, and adapts as new models and domains arrive.

Applications range from private-cloud AI platforms to autonomous scientific agents (e.g., microscopy) that retrieve domain knowledge on demand.

Representative publications (Bhattacharya et al., 2026; Saranathan et al., 2025).

References

2026

  1. ICML
    It is Time to Virtualize Foundation Models with a Self-evolving Operating System Layer
    Suparna Bhattacharya, Tarun Kumar, Cong Xu, and 7 more authors
    In International Conference on Machine Learning (ICML), 2026

2025

  1. TPC
    Advancing Autonomous Microscopy Agents with Domain-Guided Dynamic Retrieval in a Virtual Foundation Model OS
    Gayathri Saranathan, Martin Foltin, Cong Xu, and 12 more authors
    In Trillion Parameter Consortium (TPC), 2025