Quantitative research · Applied AI · Research engineering
Marcus Gawronsky is a researcher and technical founder in Cape Town, building AI systems, testing their impact and developing new ways to measure financial risk.
The work
Ideas become useful through precise models, empirical tests and reliable software.
Can an interesting idea become rigorous enough to defend and robust enough to use?
Selected work
Each project pairs a research question with the computational machinery needed to test it, reproduce it, and carry it forward.
Can company information reveal economic relationships and financial risk?
A quantitative-finance research programme connecting language-model representations, optimal transport and portfolio risk, with software for empirical tests, simulation and formal verification.
Methods: Optimal transport and Wasserstein geometry · Covariance and factor structure · Language-model representations · Reproducible research systems
Python · Lean 4 · LaTeX · Next.js
What happens when hidden financial relationships become public?
An empirical investigation of offshore entities, leaked disclosures, network structure, and market pricing, combining event studies, graph theory, and reproducible analysis.
Python · Jupyter · Econometrics · Graph theory
Search and recommendations that work from the first request.
A multimodal discovery platform connecting text and image understanding with product search and recommendations, including for visitors without interaction history.
Multimodal AI · Product discovery · Recommendation systems
Keep generated Markdown sections in sync with their source files.
A Python tool that keeps research documents connected to source files, queries and templates, with checks for stale content.
Python · Markdown · Jinja · Zensical
Research
Whether information in language-model representations can be given mathematical structure—and what that structure says about dependence, interaction, and risk.
Marcus Gawronsky · Chun-Sung Huang
The question: How does competition change firms' responses to opportunity in language-model space?
Marcus Gawronsky · Chun-Sung Huang
The question: Can distributional similarity define economically meaningful interaction fields?
Methods: Optimal transport and Wasserstein geometry · Covariance and factor structure · Language-model representations
Marcus Gawronsky · Chun-Sung Huang
The question: Can portfolio risk be bounded without estimating every cross-asset covariance directly?
Methods: Covariance and factor structure · Language-model representations · Simulation and uncertainty · Optimisation and decisions
Marcus Gawronsky · Chun-Sung Huang
The question: How much systematic covariance can be recovered from the geometry of probability-valued representations?
Methods: Optimal transport and Wasserstein geometry · Covariance and factor structure · Language-model representations
Ask about this work
Writing
The machinery around the research: reproducibility, provenance, modelling choices, and treating analysis as software.
Mega-IPOs, the convergence of public and private markets, and the case for a new approach to financial measurement.
AI in finance plays three distinct roles — a tool for financial practice, a source of economic change, and a means of measurement — and each demands different evidence.
Language models read individual companies remarkably well; the harder question, as Markowitz taught finance, is what that information means together inside a portfolio.
