MK.
MK
AI

MOVE YOUR POINTER · SELECT A SYSTEM

Prologue

MARK KAMUYA · AI + MATH · MINERVA ʼ28

AI systems,
mathematics &
software engineering.

A math, AI, and software engineering enthusiast, always on the lookout for challenging problems worth solving.

BEGIN THE STORY ↓
5MSEARCHES
644+GAMES
311RECORDS
TOP 3%MATH OLYMPIAD

01 / THE WORK

Three systems.
Three ways to make decisions.

01

MARKET DESIGN · SIMULATION

PROJECT 01

AdMechanica

Millions of individual decisions can create one complex market.

AdMechanica is an event-driven sponsored-search simulator. It lets bidding strategies compete under the same traffic and market conditions, making their behavior directly comparable.

KEY PROBLEM

Bidding strategies are difficult to compare when traffic, auctions, clicks, budgets, and conversions vary between runs.

APPROACH

Built an event-driven market with modular auction, click, and budget-pacing engines. Deterministic replay keeps 5M-search experiments reproducible.

  • Deterministic replay
  • Configurable traffic
  • Modular auction engine
  • Budget pacing
SEE PROJECT ↗
5Msearches simulated
02

GAME AI · AGENT EVALUATION

PROJECT 02

PokéMind

A strong game-playing agent must work against opponents it has never seen.

PokéMind is a tactical Pokémon TCG agent. I tested it across replay-derived decks and paired games to find strategies that generalize instead of memorizing a narrow matchup.

KEY PROBLEM

A policy that performs well against familiar decks can fail when the opponent or matchup changes.

APPROACH

Built a reproducible agent-evaluation pipeline with replay-derived matchups, counterfactual action branching, grouped ranking, seeded paired tests, and explicit promotion gates.

  • 910+ peak public score
  • Seeded paired gates
  • Counterfactual action values
  • Documented failed experiments
SEE PROJECT ↗
644+paired games evaluated
03

COMPUTATIONAL GEOMETRY · OPEN RESEARCH

PROJECT 03

Triangle Packing Atlas

A useful map shows what is verified—and where knowledge ends.

Triangle Packing Atlas is a living, reproducible map of packing knowledge across triangle shape and container geometry. It separates verified constructions, best-known records, proven controls, and explicitly open regions.

KEY PROBLEM

Computational packing results are easy to overstate: a dense picture is not a proof, and an algorithmic candidate is not necessarily optimal.

APPROACH

Built a phase-map interface, versioned coordinate dataset, independent JavaScript and Python verifiers, deterministic solver portfolio, and contribution pipeline that preserves provenance and uncertainty.

  • Dual-language verification
  • Versioned research dataset
  • Evidence-graded claims
  • Open-problem board
SEE PROJECT ↗
311independently verified records

02 / EXPERIENCE

Research.
Teaching.
Engineering.

AI Instructor

Lawrence Hall of Science, UC Berkeley

Designed and taught Python and ML concepts to 120+ high school students, then mentored over 25 teams for 200+ hours as they built voice recognition, image detection, and classification projects.

AI Research Intern

Minerva University

Built a Flask and JavaScript evaluator for enterprise AI systems, translating six governance facets into evidence requirements, validity checks, weighted controls, non-compensatory scoring, and real-time guardrail reports.

Software Engineering Intern

Stealth

Engineered reusable React Hooks and SEO components across 10+ pages, added automated accessibility tests, and shipped a lazy-loaded team experience that increased engagement by 12%.

03 / RECOGNITION

Curiosity,
unconstrained.

Awards and recognition received along the way.

01

National Math Olympiad

TOP 3% NATIONWIDE
02

Project Euler

TOP 1% · 100+ SOLVED
03

Hack the Interview

COMPETITION WINNER

04 / REACH OUT

Have a hard problem?
Let’s talk.

Open to AI/ML, mathematics, and software engineering opportunities, as well as collaborations on ambitious technical problems.

START A CONVERSATION ↗
GITHUB ↗LINKEDIN ↗