Trustworthy AI for Business

Pricing, finance, market design, generative AI, and agentic AI

This stream builds AI systems that businesses can rely on, and applies them to decisions in finance, accounting, economics, and operations. These are settings that demand learning under uncertainty, structured high-dimensional information, adaptation to heterogeneity, and operation under latency, strategic interaction, and operational constraints. Two connected directions. The first is reliable AI systems: language models whose answers are invariant to logically equivalent rephrasing, interpretable signatures of reasoning inside LLMs, uncertainty quantification for generative models and tensor-structured networks, and agentic workflows with provable convergence and latency guarantees. The second is AI for business decisions: dynamic pricing and assortment under cross-market heterogeneity, market design and extractable value in financial markets, evidence-grounded audit-risk reasoning from SEC filings, and decision systems for economics and trade. The two directions share one premise: a model used for consequential decisions should be consistent, calibrated, and fast, not only accurate on a benchmark.

Representative work: Building Logic-invariant LLMs through Logical Structure (NeurIPS, 2026), Latency-Aware Multi-Agent Workflow Learning (NeurIPS, 2026), Learning to Hand Off: Provably Convergent Workflow Learning (2026), Small-World Connectivity as a Signature of Reasoning Performance (2026), Quantifying Epistemic Uncertainty in Diffusion Models (AISTATS, 2026), Seeing Through the Brain (ICLR 2026, Oral); VERA-8B: Evidence-Grounded Audit Risk Reasoning from SEC Filings (2026), Dynamic Contextual Pricing with Doubly Non-Parametric Random Utility Models (Management Science, 2026), Transfer Learning for Contextual Joint Assortment-Pricing (2026), Maximal Extractable Value in Batch Auctions (ACM EC, 2025), ACT-Tensor (ICAIF, 2025). See publications for the full list.

Posts