Georgia Tech Byte Fight · 2026
Byte Fight 2026
I built a CPU-constrained game bot combining parallel tree search, selective expansion, hashing, vectorization, and automated evaluator tuning.
Expanded technical stack
Overview
Byte Fight imposes strict CPU limits, so I treated representation, pruning, caching, and evaluation speed as part of the search strategy.
My contribution
I engineered the bot with multithreaded minimax, alpha-beta pruning, iterative deepening, beam search, and breadth-first search, then optimized it with Zobrist hashing, SIMD vectorization, and SPSA parameter tuning.
Challenge
I needed to search deeply enough to make strong moves while staying within strict competition CPU limits.
Technical approach
I combined parallel minimax and selective expansion with transposition hashing, vectorized evaluation, and automated SPSA tuning.
Result / outcome
Our bot placed 4th out of 154 teams under Byte Fight's strict CPU constraints. I built major portions of its C++ search and evaluation architecture, combining multithreaded minimax, alpha-beta pruning, iterative deepening, beam search, breadth-first search, Zobrist hashing, SIMD-vectorized evaluation, and SPSA automated parameter tuning. Pruning reduced search work, Zobrist hashing and caching reused computations, SIMD accelerated evaluation, and multithreading increased compute utilization. Automated SPSA tuning and benchmarking enabled rapid, evidence-driven performance improvements, producing a highly efficient game-playing AI under tight compute constraints.