Embodied AI Startup Founder/ Independent AI researcher working on behavioral evaluation, epistemic robustness, and comprehension-gated intelligence systems.
My work studies how large language models fail under compression, adversarial pressure, and ethical stress.
Rather than optimizing for surface accuracy, I investigate structural robustness:
- Do models preserve constraints under ambiguity?
- Can they reject plausible falsehoods?
- Do they adapt behaviorally under pressure?
- Can capability growth be gated by demonstrated understanding?
The research arc progresses across four evaluation frameworks:
CDCT → DDFT → AGT → Comprehension-Gated Capability Growth
Separates constraint compliance from semantic accuracy under prompt compression.
arXiv: [https://arxiv.org/abs/2512.17920]
Stress-tests epistemic robustness via progressive compression and adversarial fabrication.
Under peer review. arXiv: [https://arxiv.org/abs/2512.23850]
Behavioral diagnostic distinguishing performative ethical reasoning from genuine adaptability.
Zenodo: [https://zenodo.org/records/18282166]
Recursive causality in Indian philosophical literature.
Under peer review, Zenodo: [https://zenodo.org/records/17905985]
- Behavioral Evaluation Theory
- Agentic verification systems
- Comprehension-gated training architectures
- Robustness under compression and adversarial drift
Contact: rbaxi@alumni.cmu.edu