# AI Training Manager: Bounded Closed-Loop Control of Adaptive Training Recipes

Anjali Rao, Nikhil Kamalkumar Advani. arXiv (2026).

An asynchronous LLM supervisor observes training telemetry and proposes constrained changes to a running training recipe, with experiments on deliberately induced supervised-learning and reinforcement-learning failure modes.

Studies an automated controller over active training, complementing Interactive Training’s reusable infrastructure for human and automated controllers.

[Original paper](https://arxiv.org/abs/2606.29871)

## Citation

```bibtex
@misc{rao2026aitrainingmanager,
  title = {{AI Training Manager: Bounded Closed-Loop Control of Adaptive Training Recipes}},
  author = {Rao, Anjali and Advani, Nikhil Kamalkumar},
  year = {2026},
  url = {https://arxiv.org/abs/2606.29871},
  eprint = {2606.29871},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2606.29871}
}
```