Six independent sub-scores. The exact bullets dragging you down. The exact keywords the JD asked for that you forgot to mention. All deterministic, all repeatable, and all free.
We re-extract text from your file using the same libraries the major ATS engines use (Tika, pdfplumber, mammoth). If text is rasterized inside an image, hidden behind an OCR layer, or split across columns the parser can't follow, this score drops and we tell you which page.
We run your text through a curated skill graph with thousands of synonyms (ML to machine learning, k8s to kubernetes, typos like 'tensorlfow') so legitimate skills aren't missed because of spelling. The score shows which required skills are present, which preferred skills you mention, and which the JD asked for that are absent.
We pull each role's date range, compute total relevant experience, and compare it against the minimum years parsed from the JD. Senior bullet density is sanity-checked against title to flag the 'principal engineer who has six months of work history' anti-pattern.
Bachelor, master, PhD are normalized, accreditation hints surfaced, and we never penalize bootcamp grads when the JD says 'or equivalent experience'.
Strong CVs have numbers in 55-75% of bullets. Below that and the AI rewrite suggestion engine starts proposing measurable outcomes for bullets that currently end in 'and made improvements'.
Surprising how many CVs lose 8 points to an unparseable header. We verify that the email, phone, and city all live in the first 150 characters of the document, the order most ATS regexes expect.
Most ATS checkers give you a single number. We do too, but the number is just the arithmetic mean of the six sub-scores. The verdict bands are calibrated against tens of thousands of recruiter inboxes:
Yes. Upload a PDF or DOCX, get all six sub-scores, see which lines dragged you down, and export the result. No paywall on the score itself.
First score in under 60 seconds for a typical 1 or 2 page PDF. Larger documents (8+ page academic CVs) take 2 to 3 seconds longer because we parse every page through the full library set.
The text-extraction layer covers what Taleo, Workday, Greenhouse, Lever, iCIMS, and Ashby use under the hood (pdf text extraction, DOCX traversal, fallback OCR detection). The scoring layer uses a deterministic skill graph rather than a black-box LLM, so the same CV always gets the same score.
PDF or DOCX are the most reliable. Pasted plain text works but loses formatting signals (bold section headers, font sizes) that the parseability sub-score uses. A scanned image PDF will score low on parseability until you OCR it first.
85+ is strong (most resumes from senior candidates land here once tuned). 70 to 85 is good. 55 to 70 is borderline (you'll likely lose the keyword-only filter step). Below 55 means a parser will struggle, and you should fix that before tailoring to any JD.
Yes. Tables, multi-column layouts that don't preserve reading order, text-in-images, custom section names ('My journey'), unusual fonts that don't embed, and image-only PDFs all surface as specific parseability warnings.
No. The score is computed deterministically by our own parsers, not by an LLM. Only the optional AI rewrite step calls Claude, and only on the snippet you select.
Once you've fixed parseability, line up the score against a specific posting.
Rewrite the bullets the checker flagged. Keep your tone, land the keywords.
Once the score is high, swap to a stronger-looking template without losing the score.
Upload a PDF or DOCX. See every line the parser misses.
Check my resume →