Marathi, an Indo-Aryan language predominantly spoken in India, holds a rich linguistic and cultural legacy. However, the rise of code-mixed and script-mixed content, especially on social media, poses significant challenges for automated translation. These challenges primarily stem from inconsistent spellings, the use of slang, and the frequent mixing of Roman Marathi with English, which complicates standard machine translation processes. This paper explores the difficulties in transliterating Marathi code-mixed and script-mixed text, where Roman Marathi words often coexist with English. Current transliteration tools often struggle with non-standard spellings and regionally influenced variations, particularly when converting Roman script to Devanagari for languages like Marathi. To address these limitations, this study introduces a validation mechanism augmented by a manually curated repository of phonetic variants. This resource includes informal spellings and alternative phonetic representations commonly found in user-generated content. The proposed approach centers on two key components: phonetic normalization, which standardizes diverse Romanized inputs, and the manual compilation of phonetic variation sets, designed to capture the irregularities of informal language use better. These strategies are crucial for enhancing transliteration accuracy in real-world contexts, where deviations from standard linguistic conventions are common. By addressing these challenges, the study enhances the reliability and adaptability of transliteration systems, contributing to broader applications in natural language processing. The proposed approach ultimately seeks to enhance the quality and effectiveness of Code-mixed Marathi transliteration in contemporary digital communication.