DcisionAI Research

Research

For forty years, optimization lived in PhD silos, too expensive to model, too brittle to deploy, too opaque to trust. Generative AI gave every team a writer. It did not give them a decider.

Our research is an evolving argument: that the next infrastructure layer of the enterprise is not another dashboard or another chatbot. It is a decision layer that runs one loop — CORE (Context, Options, Resolve, Encode) — solver-grounded, constraint-aware, and auditable by default. One engine, reached through the platform, MCP, or agent-to-agent.

The argument in brief, read the thesis →

PapersFormal · technical
Working Paper2026

Infeasibility as Diagnostic: IIS Decomposition for Auditing Business Rules in Portfolio Optimization

DcisionAI Research

Infeasibility is conventionally treated as a solver failure. We argue the opposite: an infeasible model is the most valuable diagnostic an optimization system can produce. Using Irreducible Infeasible Subsystem (IIS) decomposition combined with shadow-price analysis, we show how conflict detection surfaces business-rule contradictions, ESG mandates against minimum-yield floors, sector caps against liquidity requirements, that humans cannot reliably enumerate. We present a case study on a $240M multi-asset portfolio where IIS resolution exposed a $2.1M latent tradeoff and triggered a documented, auditable human override.

OptimizationConstraint ProgrammingPortfolio Management
PublicationsEssays and articles

The argument is unfinished by design. Each new deployment, each surfaced infeasibility, each documented override sharpens the argument. New publications land here.