About Joseph Bunster
About Joe Bunster
I’m a quantitative researcher and modeler working across AI evaluation and financial model risk. I build models, examine their assumptions, and turn findings into evidence for research and business decisions.
Professional experience
Meta
Research Collaborator V · Generative AI
I evaluate frontier LLM reasoning, prompts, rubrics, and reference solutions across mathematics and STEM. I partner with researchers on research questions, evaluation workflows, and vendor strategy, and lead early-stage research on annotation diagnosis, repair, and verification.
Within a four-person leadership group, I advise three external data vendors, translating research priorities and evaluation findings into recommendations for annotation workflows and quality improvement.
JPMorgan Chase
Quantitative Analyst, Senior Associate (2023–2026)
Quantitative Analyst, Associate (2021–2023)
I built Python challenger models and led independent model reviews across financial applications. I assessed statistical robustness, challenged hidden assumptions, and stress-tested model behavior across market regimes to identify weaknesses. I also examined whether the models’ econometric relationships were supported by sound economic reasoning.
I led the CIB Finance Model Risk team’s frontier LLM initiative, applying supervised fine-tuning and domain-specific evaluation to model risk workflows.
I developed guidance for an LLM-assisted document-readiness workflow that gave model developers automated feedback on potential critical documentation gaps before formal review, helping reduce back-and-forth with reviewers.
How I like to work
I start with the decision a model is meant to support, make its assumptions explicit, and design checks that could reveal where it fails. I move between mathematical reasoning, Python implementation, and communication with researchers and model owners.
As founder of Regime Alpha and HomeOnCue, I own the problem definition, technical choices, and product direction. I use AI to accelerate development while remaining responsible for the code, assumptions, and validation. I’m interested in senior research roles and focused consulting projects where I can carry that responsibility from an initial question through a tested result.
What interests me
I’m interested in how we model uncertainty, how assumptions shape the answers we get, and how mathematical ideas become useful tools for decisions.
My interests include optimal transport, optimal control, probability, stochastic modeling, time-series analysis, and machine learning. My thesis projects explored financial forecasting and retirement portfolio decisions. Today, those questions inform my work on portfolio regimes and AI reasoning evaluation.
Academic background
NYU Courant Institute
M.S. Mathematics
Master’s thesis, 2022: Optimal Transport and Dynamic Programming in Financial Time Series Analysis
Farmingdale State College
B.S. Applied Mathematics
Undergraduate thesis, 2018: Optimal Control Theory in Retirement Portfolio Management