Unlock: Reinforcement Learning for Synthesis Planning

Retrosynthesis as tree search: AlphaChem-style MCTS over learned reaction templates, transformer-based template-free models, and reward shaping with synthetic accessibility heuristics.