Problem
Career advice for students is mostly vibes: “learn cloud”, “get certs”. The actually useful question (which specific skills separate my résumé from the jobs posted in my region right now?) has a data answer, but nobody was computing it.
Approach
- Scraped and aggregated regional job postings into a structured dataset of in-demand titles, required skills, and qualifications.
- Parsed uploaded résumés and matched their content against the dataset: not a generic skills taxonomy, but what employers in the region were literally asking for that month.
- Surfaced the result as a gap report: the skills you already demonstrate, the ones you’re missing for your target role, and tailored role recommendations where your existing profile is already competitive.
What I learned
The scraping was the easy half. The real work was normalization: a hundred job postings say “communication skills” ninety different ways, and skill matching is only as good as the entity resolution underneath it. This is the project where I stopped treating messy text as a regex problem and started treating it as an NLP problem. The regexes are still in the codebase. They know what they did.