This paper presents Meridian 0.1, a data-first approach to academic proofreading using a LaTeX-aware small language model (SLM). The study investigates how training-data design, task decomposition, and document-structure awareness influence academic proofreading performance. Meridian-AC-Nano is introduced as a 751.6M-parameter multi-task academic assistant designed around structured academic text and LaTeX-aware processing. The work focuses on the hypothesis that improvements in data design and task formulation can provide substantial gains without relying exclusively on larger model architectures. The paper describes the system design, dataset strategy, training methodology, evaluation framework, and limitations of the approach.