CV CheckerCV Checker

Sole engineer: product design, the scoring engine, the AI layer, the bilingual interface and the security model.

Most resume checkers return a single opaque number and a generic AI rewrite. Candidates cannot tell why they scored what they scored, and AI-written cover letters happily invent achievements the resume never claims. Many tools also keep the uploaded CVs.

A Next.js app that separates scoring from generation. The ATS score comes from pure TypeScript rules (parseability, contact details, sections, role alignment, hard skills, experience, keywords and achievement quality), each shown with its signal, weight and points, so the whole score can be audited.

A resume health audit checks contact completeness, timeline order, employment gaps to the month, "ghost skills" listed without evidence, quantified achievements, length and density. With a job post, every requirement is classified as required or preferred and mapped to the exact resume lines that support it, with an evidence-strength rating and a seniority check. The cover letter generator (job-tailored or general) only uses verified achievements, and the whole analysis exports to a PDF report.

Uploads (PDF, DOCX or TXT up to 5 MB) are validated by file signature, MIME type and extension, then parsed with pdf.js or Mammoth, with scanned-file detection. A deterministic engine computes the score, health checks and matching. Gemini is used only for structured extraction and writing; its JSON output is validated with Zod and checked against the resume text, and a deterministic extractor keeps the app fully working when the model is unavailable. PDF reports are rendered with headless Chromium.

Keeping the language model honest: anything it returns that is not present in the resume is rejected before it reaches the user. Explaining a score instead of just showing it, and detecting gaps and unsupported skills reliably across many resume layouts and both languages.

Deterministic rules for anything that is a judgement about the candidate, and the model only for extraction and wording. A full fallback path so a model outage never breaks an analysis. Privacy by design: resumes are processed per request and never written to a database, disk, browser storage or logs.

On a realistic test resume checked against a Senior Full-Stack job post, the app returned an ATS compatibility of 80/100 with 95% confidence, resume health of 83/100 and a 73% job match. It found the 7-month gap between two jobs (Aug 2021 to Mar 2022), counted 6 quantified achievements, flagged Docker as a missing must-have and PostgreSQL as a missing nice-to-have, and mapped 12 requirements to resume evidence. A job-tailored cover letter was generated in about 3.3 seconds using only the resume's real numbers.