Amit Sheth

Standard NLP playbooks fail under real-world clinical constraints: across 300 controlled experiments evaluating 31 popular modeling tricks, only 5 provided reliable gains over task-grounded baselines.

LLMs often prioritize safety, leading to conservative predictions about recipe suitability for diabetes, but some models can significantly outperform others when reasoning with dietary guidelines.

Knowledge infusion in generative models can be strategically layered to achieve a dramatic 70.97% reduction in knowledge-violating outputs.

Achieve near-perfect success (98%+) in real-time causal diagnostics for smart manufacturing with a neurosymbolic multi-agent copilot, proving the viability of interpretable AI in complex industrial settings.