LLMs Uncover Soft Skills in ML Engineering CVs Beyond Keywords.

While soft skills are widely recognized as crucial for collaboration among ML engineers, data scientists, and software engineers, current understanding largely stems from employer perspectives found in job ads and interviews. Little is known about how candidates themselves articulate these competencies on their CVs, and existing CV-mining tools are often keyword-based, missing skills conveyed through narrative. This new research addresses these gaps by analyzing a balanced corpus of 300 curated CVs from these three roles. The study employs an LLM-based pipeline, validated against human annotations, to extract both explicitly listed and implicitly narrated soft skills. This approach allows for a more comprehensive understanding than traditional keyword-based methods. Researchers then tested 13 hypotheses derived from demand-side literature regarding role signatures, seniority progression, and disclosure styles. The findings reveal that candidates predominantly communicate soft skills through narrative descriptions rather than simple keyword lists, with a ratio of approximately three to one. This narrative approach is even more pronounced for highly valued competencies like leadership, coordination, and mentoring (88-96% narrative). Seniority significantly increases the likelihood of articulating leadership skills. Contrary to some assumptions, software engineers articulate leadership at half the rate of their peers. The study concludes that technical candidates do articulate soft skills, but keyword-based screening methods systematically overlook a significant portion of this information.