AI-Powered Job Matching Board
The problem
The recruitment team spent hundreds of hours reviewing unstructured CVs to find the right profile for technical vacancies.
The solution
Smart job board (for LTC) that parses and extracts data from uploaded CVs, using AI to perform an automatic match between the candidate's skills and open vacancies.
Architecture
Next.js front end. A Python NLP service parses unstructured CVs into a profile. Matching scores skills against vacancies. AWS hosts compute. The candidate applies without filling the same form the parser just read.
How it was measured
Figure reported by the matching engineer, on LTC's initial screening flow. The 75% compares linear reading of unstructured CVs with the ranking the model produces. The engagement is recorded as 2024–2025; there is no public start month. No third-party audit.
How do you shorten screening of technical CVs without an endless form?
The candidate uploads the CV. A Python service extracts skills; matching scores them against open vacancies in Next.js on AWS. The recruiter starts from a ranking, not an unstructured pile.
The work
LTC had a team reading unstructured technical CVs. Hundreds of hours of initial screening: open PDF, hunt for a stack, reject. On the other side, the candidate retyped what the PDF already said. The board I built parses the file, builds a profile, and scores the match against open vacancies. The candidate does not fill the same form the parser just read.
Next.js is the face. Python does the NLP. AWS hosts. Matching is not a magical “culture” embedding. It is skills against a technical vacancy's requirements. PDFs arrive dirty: columns, tables, headers. A parser that fails silently is worse than a human, so the recruiter sees the ranking and the file, not a verdict. The 75% is initial screening time the operator measured between the linear pile and the ranking. It is not time-to-hire, not hire quality, and there is no HR study attached.
Public dates are 2024–2025, with no start month. I would rather publish an honest year than an invented month. LTC appears in the original portfolio copy as the board's recipient; it is not a listed job in this site's experience section, and I do not dress it as one. Anyone looking for Towel or UVP on this page has the wrong URL.
Relation to the rest of the record: UVP's RAG is also retrieve-and-rank over documents. Here the document is a CV and the index is a vacancy. It is not the same product. It is the same refusal to make a human the parser. The screenshots are that board, not a generic job-board template.
Screenshots
