Data Analyst · MS Business Analytics, University of California Irvine
I approach analytical problems through formal quantitative frameworks and deploy understanding through project work. My focus is applying rigorous methods — optimization, game theory, statistical modeling — to domains where structured analysis is underused.
Currently building fluency in Python and ML methods through direct project deployment rather than coursework alone. I learn by building something with the tool until the tool becomes transparent and the problem becomes the subject.
PoE2 Armour Optimization & Survivability Analysis
Formal derivation of the PoE2 armour damage-reduction formula, identifying the optimal 12–20× armour-to-life corridor and demonstrating that physical-to-elemental conversion mechanics change the calculus structurally. Includes a mathematical proof that 50% conversion is exactly 2× more mitigation-efficient than doubling armour — independent of hit size and armour value — and a full cross-archetype survivability comparison across Shaman, Smith of Kitava, and Witchhunter.
Delivered as a standalone interactive browser tool: character setup, damage simulation with conversion model, Witchhunter Sorcery Ward calculator, and a full DR reference matrix.
game-theory optimization quantitative-modeling interactive-html
Two-Phase Binned Parameter Search
An algorithm for identifying optimal k-values on monotonic objective functions without exhaustive search. Two-phase coarse-to-fine bin traversal locates the optimal region, then refines to a provably near-optimal k̂ estimate. Validated both theoretically (convergence guarantees) and empirically — reduces evaluation cost substantially relative to full enumeration while preserving solution quality bounds.
algorithms parameter-optimization python ML
- Market research & pricing strategy
- Game theory & operations research
- Process & workflow optimization
- Applied ML & data science methods
- AI as general-purpose infrastructure
- Decision modeling under uncertainty
Python SQL pandas scikit-learn quantitative modeling optimization theory statistical analysis LaTeX data visualization
The correct response to AI's energy cost is clean energy at the infrastructure layer — not restraint from use.
I treat AI as a general-purpose technology: applicable across domains, with incremental computation costs on a long-run downward trajectory as the energy and cooling infrastructure matures. The sustainability question is real, but the solution lives at the generation and cooling layer — carbon-free energy and water-efficient cooling at scale — not in individual usage decisions made while that infrastructure is still being built.
What actually limits value extraction from AI is not access to the models — it is literacy and analytical structuring ability. The value of any human-AI collaboration scales directly with the operator's capacity for project decomposition, logical scaffolding, and moving fluidly between the micro and macro view of a problem. These are the binding constraints, and they are the ones worth investing in. Education and AI literacy are not gatekeepers in the exclusionary sense — they are amplifiers: the better you are at breaking down problems, the more work the tool can do.
UC Irvine MSBA · denarot@uci.edu · github.com/denarot