Skip to content

Research

Evaluating how decision frameworks perform on identical history.

My research asks one question: when different decision frameworks are evaluated on identical historical information, how do they actually perform? Answering it rigorously means building the simulation and evaluation infrastructure to replay decades of history under each framework and score them the same way. This is active Honors work at Oregon State University.

Oregon State University Honors CollegeActive research

Honors thesis

Building infrastructure capable of evaluating decades of market behavior under multiple decision frameworks.

The same historical information, replayed under every framework, scored on equal footing. The contribution is the methodology and the infrastructure — a controlled environment where human, quantitative, and AI-driven decision systems can be compared rigorously rather than anecdotally.

Evaluation at scale

Decades of history, replayed at scale.

Each framework is evaluated against decades of historical market behavior, replayed repeatedly under identical conditions. The research workflows are actively being designed and built around Oregon State University's expanding AI computing infrastructure — including NVIDIA-backed research resources — to run large-scale simulation and evaluation workloads.

Decades
of historical data replayed
4
competing decision frameworks
Repeated
large-scale evaluation runs
NVIDIA
research compute (in design)
Quantitative systemsHuman decision systemsHuman + AI systemsAI-first systems

The experiment

Decision frameworks under evaluation

The same market history, replayed under four distinct decision frameworks — evaluated on equal footing.

01

Human decision making

Discretionary judgment as a baseline

02

Traditional quantitative models

Rules-based, statistical frameworks

03

Human + AI systems

AI augments, humans decide

04

AI-first systems

Model-driven decision processes

Where the rigor lives

A methodological program

The contribution is in how the comparison is designed, validated, and evaluated — built to be reproducible.

Research methodology
Experimental design
Statistical validation
Decision systems
Model evaluation
Simulation science

The engine

Simulation infrastructure

At the core is a simulation engine: a controlled, reproducible environment that replays historical information and lets each decision framework act on exactly the same data. It's built framework-agnostic, so a new decision system can be dropped in and evaluated under identical conditions.

Historical replay engine

Decades of data, deterministic playback

Controlled environment

Identical information for every framework

Reproducible runs

Same inputs, same evaluation, every time

Framework-agnostic harness

Drop in a new decision system and score it

Research areas

Applied mathematics & machine learning systems

The work sits at the intersection of quantitative methods, machine learning, simulation, and evaluation.

01

Quantitative finance

Rules-based and statistical frameworks as a rigorous comparison baseline.

02

Machine learning

Applied ML systems for decision modeling and evaluation.

03

Large language models

LLM-driven reasoning and decision workflows under structured evaluation.

04

Historical simulation

Replaying decades of market behavior as a controlled evaluation environment.

05

Human + AI decision systems

How augmentation changes outcomes versus humans or models alone.

06

Statistical evaluation

Methodology for comparing frameworks fairly and reproducibly.

07

Market behavior analysis

Characterizing regimes and dynamics that frameworks must hold up against.

Applied research

Research that ships into real software.

Many of the ideas explored in the research program are already reflected in real software through Agentic CRM, where AI recommendations remain subject to human approval and measurable outcomes. Research supports the engineering; the engineering validates the research.

Human-in-the-loop systemsAI recommendationsDecision support systemsEvaluation frameworksControlled AI autonomy
See it in Agentic CRM
Research
supports engineering
Engineering
validates research

Let's compare notes.

If you work on AI-assisted decision systems, simulation infrastructure, or applied ML evaluation, I'd be glad to talk.