Arman Boyaci

I build mathematical models to solve complex planning problems involving conflicting objectives and high uncertainty.

I've applied these models across Supply Chain (replenishment, routing, capacity planning), Finance (investment strategies), and Marketing (pricing & promotion planning).

I focus on delivering transparent, reliable AI solutions that drive measurable business results.

✎ Technical Notes

Machine Learning

• Exploration vs Exploration Exploration is essential in machine learning settings where our decisions dictate the data we observe.
• Log Transformation Bias Applying log transformation to the output variable introduces a bias.
• Modeling Seasonality Indicator variables are easy to interpret, and Fourier series are smooth; periodic splines offer the best of both.
• Monotonicity We explore ways to constrain coefficient values to be non-decreasing.
• Non-Negativity Constraints What are my options for enforcing some of the coefficients to be non-negative?

Playing Games

• Tic-Tac-Toe There are 12 unique ways to start. In 7 of them, the first player can always win with the right moves.
• Blackjack The house always wins because if both the player and the dealer bust, the dealer still keeps the money.
• Guess Who? The best move is to ask a question that splits the current board into two as evenly as possible.
• Pop-it The player who breaks the symmetry loses; the opponent wins by restoring it.
• Dots and Boxes On a 3×3 grid, the first player can always win by following a specific strategy which involves a sacrifice.

Solving Puzzles

• Sudoku We present graph-theoretical interpretations of some popular Sudoku solving methods.
• Peg Solitaire We present a method that forms a heart shape followed by a T-shape, making it easier to complete the puzzle.
• 15 Puzzle A simple solution method involves carefully rotating 2×2 or 3×3 blocks.
• Tower of Hanoi You can solve the puzzle by moving odd-numbered disks left and even-numbered disks right.