Publications

Olanrewaju Muili, Hassan A. Babaie. Sinkhole susceptibility analysis using machine learning for west central Florida. Applied Computing and Geosciences, vol. 27, 100262, 2025.

Peer-reviewed machine-learning study of sinkhole susceptibility in west central Florida — geospatial feature engineering and supervised models applied to karst hazard mapping. The modeling approach later informed the Karst Intelligence Agent decision-support system.

[Paper][DOI][Code]

@article{muili2025sinkhole,
  author  = {Muili, Olanrewaju and Babaie, Hassan A.},
  title   = {Sinkhole susceptibility analysis using machine learning for west central Florida},
  journal = {Applied Computing and Geosciences},
  volume  = {27},
  pages   = {100262},
  year    = {2025},
  doi     = {10.1016/j.acags.2025.100262}
}

Olanrewaju Muili. Evaluating hybrid guardrail architectures for prompt injection defense in large language models. SSRN preprint, 2026.

Sole-authored systematic evaluation of baseline, regex-only, and hybrid (regex + LLM classifier) guardrails on a two-tier 625-prompt benchmark spanning 18 attack categories. On the adversarially hard tier, the hybrid configuration reaches a 98.9% attack block rate (recall) with a 0.0% benign false-positive rate (F1 = 0.994), while regex-only filtering degrades to a 24.6% block rate.

[Paper][DOI][PDF][Project]

@article{muili2026guardrails,
  author = {Muili, Olanrewaju},
  title  = {Evaluating Hybrid Guardrail Architectures for Prompt Injection Defense in Large Language Models},
  year   = {2026},
  note   = {SSRN preprint},
  doi    = {10.2139/ssrn.6246379}
}

Olanrewaju Muili. Learning When To Trust: Training-Time Calibration for Reinforcement Learning under Adversarial Observation Corruption. Manuscript in preparation.

Investigates whether an agent can learn to estimate the reliability of potentially corrupted observations and adapt its sequential decision-making accordingly. Experiments are being developed in the open on TraceVox Research.

[Project]