HOW THIS WORKS

No model reads your resume.

Your resume is graded by a set of rules you can read, which give the same answer every time you ask, and nothing you upload is kept. That is worth saying plainly, because the usual alternative is a language model, generally someone else's, in another country, under a policy that reserves the right to pass your details on. There is no model in this code path at all.

What actually leaves your browser

What you do	What is sent	What is kept
Check a resume	The file, over HTTPS, to our own server in a single request	Nothing. The bytes are parsed in memory and dropped when the response is written
Check a Naukri or LinkedIn profile	The text you pasted	Nothing
Track applications	Nothing at all. The tracker has no network call in it	Your own browser's local storage, on your device
Check a government photo or signature	Nothing at all. The image is decoded and measured in the page	Nothing - it never reaches a server
Sign in and save a score	Your email and the numeric score	The score history, until you delete it

No third-party analytics, no advertising pixels, no session recording, and no resume content in our logs. We are a data fiduciary under India's Digital Personal Data Protection Act, and the simplest way to honour that is to hold almost nothing.

Why not just use an LLM?

Because it would have to see your resume

A resume is a dossier: full name, phone number, home city, employer history, salary signals, sometimes a photograph and a date of birth. Sending it to a model provider means a third party now holds it. Published privacy policies in this category are explicit that user input is shared with the AI providers they call.

Because the answer would move

Ask a model the same question twice and you can get two answers. A checker whose score changes on re-upload is not measuring anything. These rules are deterministic: same resume, same score, today and next year.

Because an ATS is not intelligent either

The thing you are being filtered by is a parser and a keyword index. Modelling it with a language model is using the wrong instrument - it will fluently describe a resume that a parser cannot read at all.

Because it invites the wrong fix

Generative tools rewrite the document for you. That produces a resume you cannot defend in the interview, in a register recruiters have learned to spot. Findings here tell you what is wrong and leave the writing to you.

What the rules actually are

Parseability - the file is opened the way a parser opens it: text layer, column order, tables, images, headers and footers. If your phone number lives in a header, it is reported as missing, because that is what the ATS sees.
Contact and section detection - regex, tolerant of the formats people actually use, including +91 98765 43210 with spaces in it.
Skills - a maintained dictionary, matched literally with an abbreviation table, so ML and Machine Learning are the same claim.
Job-description matching - the posting's title and skills against yours, reported as findings that never move your score, because a résumé is not worse for being aimed at a different job.
Evidence - a skill named only in the skills list and nowhere in your experience is flagged as a claim without evidence. That is the check that keyword-stuffing tools cannot make, because stuffing is what their score rewards.

There is no hidden model behind any of it. The scoring is arithmetic over these findings, which is why the same file always returns the same number.

The trade-off, stated plainly

Rules cannot tell you whether you are a good fit for a role, whether your bullet points are persuasive, or what to write instead. A model can attempt all three, badly and differently each time. We would rather do the mechanical part exactly right and be honest that the judgement is still yours.

Check a resume