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
- ICMLIt is Time to Virtualize Foundation Models with a Self-evolving Operating System LayerIn International Conference on Machine Learning (ICML), 2026
2025
- TPCAdvancing Autonomous Microscopy Agents with Domain-Guided Dynamic Retrieval in a Virtual Foundation Model OSIn Trillion Parameter Consortium (TPC), 2025