Building the infrastructure that tells you when AI is wrong — and working to understand why.

I design and evaluate AI systems with a focus on measurable safety, reliability, and bias reduction. My research on demographic bias in clinical AI was published in Communications Medicine (Nature Portfolio), 2026, and I currently build evaluation frameworks and agentic observability infrastructure for deployed AI systems.

Currently focused on: evaluation methodology, bias mitigation, and observability for agentic AI systems.

🔗 Portfolio · About Me · Google Scholar · LinkedIn

• Bias Drift Detection — version-tracking bias evaluation pipeline across 6 LLMs using 400 synthetic clinical vignettes; found a 4.4x reduction in demographic pain-score disparity from a prompt wording change alone.
• Vedic Mathematics for AI Acceleration — a four-phase study (Python → C++ → CUDA → transformer attention) testing whether an ancient Vedic multiplication technique improves ML performance; found a tiling strategy with 16x more concurrent computation and 19% lower memory traffic than FlashAttention.
• DeepGuard (in progress) — deepfake and AI-generated video detection using motion entropy as a computationally cheap signal, with a Claude-powered reporting layer.

📫 Best reached by email: visdav8@gmail.com