Olanrewaju Muili
PhD Researcher & Research Engineer
Trustworthy Multimodal Agents · Reinforcement Learning · Adversarial ML
Ph.D. Student in Computer Science, University of Southern Mississippi · U.S. Department of Energy Marine Energy Postgraduate Fellow
I study how autonomous AI systems can estimate when their observations are trustworthy and make reliable decisions under uncertainty, corruption, distribution shift, and adversarial conditions.
Research at a glance
- DOE
U.S. Department of Energy Marine Energy Postgraduate Fellow
Research engineering with Deep Anchor Solutions — finite-element workflows and simulation-to-ML datasets.
- 2025
Peer-reviewed ML publication
Sinkhole susceptibility via machine learning, Applied Computing and Geosciences.
- 625
prompt adversarial benchmark
Sole-authored LLM safety preprint: two-tier benchmark across 18 attack categories.
- 98.9%
recall, 0.0% false positives
Hybrid guardrail configuration on the hard (adversarial) benchmark tier.
- 5,000+
numerical simulations
Parametric finite-element runs generated for simulation-to-ML research at the DOE fellowship host.
- PyPI
Public research software
tracevox-ai: open research environment for reproducible RL and trustworthy-AI experiments.
Current research
Trust-Calibrated Reinforcement Learning for Adversarially Robust Multimodal Agents
Modern multimodal agents typically assume their observations are reliable. In realistic environments, individual modalities can become noisy, missing, corrupted, contradictory, shifted, or adversarially manipulated. My research asks whether an agent can estimate the reliability of its own observations and adapt its sequential decision-making accordingly.
Current manuscript: Learning When To Trust: Training-Time Calibration for Reinforcement Learning under Adversarial Observation Corruption — in preparation.
Ph.D. research at the University of Southern Mississippi, advised by Dr. Rabab Abdelfattah. Experiments developed in the open on TraceVox Research.
Research questions & formulation →View a recorded experiment
TraceVox
The engineering platform grew out of a broader interest in building inspectable, reproducible systems for evaluating AI behavior. TraceVox Research houses the research experiments; TraceVox AI focuses on production evaluation and observability.
TraceVox Research
An open research environment for reproducible experiments in trustworthy AI, reinforcement learning, multimodal agents, corruption robustness, adversarial robustness, evaluation, and AI safety — with recorded per-timestep traces, exact replays, and machine-readable published experiments.
TraceVox AI (Platform)
Engineering work around production AI observability, evaluation, tracing, safety monitoring, and agent/LLM infrastructure — the applied counterpart to the research environment.