Tailor your résumé to the job. Defend every line of it.
RoleVerity rewrites your résumé against one specific posting using only what your own career record can prove — then hands back a reviewed, truth-checked DOCX. No invented experience. No keyword soup.
The demo is a seeded fictional candidate. No card, nothing to install, and your career record is never used to train a model.
A general AI will happily write a résumé you have to disown by the second interview.
Senior candidates do not have a writing problem. They have a proof problem: the posting asks for nine things, your history covers seven of them under different names, and the rewrite has to stay true to what you actually did.
It writes experience you never had
“Led a team of eight” reads well until the hiring manager asks who reported to you. A general chat model has no idea which parts of your history are real, so it fills the gap with what the posting wanted to hear.
It keyword-stuffs for a machine, not a person
Terms get sprinkled in to beat an imagined filter. The document that comes back is optimized for a score nobody at the company ever sees, and reads like it.
It hands back a document you have to rebuild
Copy out, paste into a fresh template, spend forty minutes fixing spacing and page breaks, then do it again for the next role.
“Drove a 40% increase in revenue by leading the platform migration.”
Reads well. Nothing in your record says 40%, and nothing says you led it.
Two unsupported terms: “40%” and “leading”. No evidence row backs either one.
The edit stays on screen so you can see what was proposed. The accept button never lights up.
Four steps, and a document at the end of them.
- 01
Import your résumé into a Career Brain
Upload the DOCX you already use. The parser proposes draft experiences, projects, skills and evidence — each one quoting the résumé text it came from. Nothing becomes a fact until you confirm it, one card at a time. You can also type rows in by hand.
- 02
Paste the job description
One paste creates the application, the interview pipeline and a verbatim copy of the posting, and extracts the requirements as a list. Paste a URL instead and it will try to fetch the readable posting text first; boards that fight back still take a paste. Analysis then answers each extracted requirement with the evidence that supports it — and names the gaps it cannot cover.
- 03
Tailor against that posting
Resume Studio proposes bullet rewrites. Each rewrite names the requirement it serves and the evidence behind it. Requirements your record cannot answer are reported as gaps rather than papered over. If the posting has no extracted requirements, or your Career Brain has no evidence, it refuses to generate at all.
- 04
Review, then download the real file
Truth Guard classifies every proposed edit before you can take it. Accepted edits are written into your original archive — the untouched bytes stay untouched, so the document you send is still your document. Your review, decisions and generated versions survive a reload.
Three verdicts, and only two of them are yours to take.
Every proposed edit is judged against the evidence cited for it, by a model and then by a deterministic lexical re-check that can only lower the model’s verdict, never raise it. A substantive term with no evidence behind it is not a reframe — it is unsupported, and the accept button never lights up.
“Built React dashboards rendering real-time vehicle telemetry over MQTT.”
Traced to a confirmed Career Brain evidence row. Accept it.
“Owned the front-end surface pilot customers used daily.”
Same underlying evidence, angled at the posting's wording. Yours to judge.
“Led a team of eight engineers through a platform migration.”
No evidence row supports “team of eight”. Readable, and impossible to accept.
Illustrative wording. The verdicts, the citation rule and the block on accepting an unsupported edit are what the product does.
The tailoring is the point. The tracker is what makes it repeatable.
Career Brain
Experience, projects, skills and evidence as structured rows with provenance and a confidence level. Imported drafts stay separate from confirmed facts, and only confirmed facts are ever sent to a model.
Requirement-by-requirement analysis
A verdict plus one finding per extracted requirement, each carrying its evidence ids and its gaps. No score, no percentage, no invented probability of getting hired.
Application tracker and interview pipeline
Company, role, level, comp range, status and starred priority, with interview stages you can add, rename, reorder and schedule. Status and star changes write themselves into the log.
Cover letters from the same evidence
Draft against the stored requirements, the posting and the mission text you paste — sourced from the same confirmed rows the résumé work uses. Copy-and-paste output today: no saved drafts and no export yet.
Private by construction
Row-level security on every user-owned table, verified by reading as the wrong account and getting nothing. Résumés and artifacts sit in private buckets behind short-lived signed links.
Export and delete on your terms
A full JSON recovery export plus per-table CSVs, any time. Deleting the account removes your stored files and cascades your rows — no email thread required.
The cuts are the product.
Anything that would make the output easier to sell and harder to defend was left out on purpose, and is written into the build contract rather than a marketing promise.
No match percentage
A number invented by a model that has never seen the hiring bar is a lie with a decimal point. The analyses table has no score column, and it is not getting one.
No mass-apply
Nothing here fires applications at hundreds of postings, scrapes job boards at scale or emails employers as you. One posting at a time, reviewed by you.
No unlimited AI
Paid access sells a fixed number of completed application packages, not a bottomless token budget nobody can price honestly.
No silent AI writes
Extraction fills fields you can edit. Import drafts wait for confirmation. Résumé edits reach the file only when you accept them.
Built by one operator, in public
I am David Shin, and I built RoleVerity while running my own search, because every AI résumé tool I tried was willing to write things about me that were not true. This one is not. It is early and I would rather tell you where it stops: checkout, hosted email, PDF output, cover-letter export and error monitoring are not built yet. The evidence loop — import, analyze, tailor, truth-check, download a real DOCX — is.
If you are a mid-career or senior candidate who has to survive the interview after the résumé gets you in the room, the demo is the fastest way to judge it. Then write to me and tell me what broke.
David Shin · hello@roleverity.app
You pay for finished application packages, never for unlimited anything.
Self-serve checkout is not live yet. The free tier and the demo work today; the founding cohort is arranged and invoiced directly, and the pass prices below are what we intend to ship.
Free
See the whole loop on your own material.
- Application tracker and interview pipeline
- Career Brain, by import or by hand
- Job analysis against your evidence
- One complete verified export
Founding cohort
Open nowA bounded number of verified packages, plus direct help from the person who built it.
- Hands-on setup of your Career Brain
- A stated package allowance over a stated window
- A direct line to the operator, not a ticket queue
- Refundable within 14 days if you have not downloaded a package
Search pass
Self-serve, non-renewing, capped in packages rather than tokens.
- A fixed number of completed application packages
- A 90-day pass is planned at $79
- Expires instead of quietly renewing
- Checkout is not live yet — these are the numbers we intend to ship
Refund terms live on the refunds page and apply the moment you pay.
The questions that decide it.
How is this different from asking ChatGPT?
ChatGPT has no idea which parts of your history are real, so when the posting asks for something you did not do, it writes it anyway. RoleVerity reads only from evidence you confirmed, cites that evidence on every rewrite, blocks the edits it cannot source, and edits your actual DOCX instead of handing back text to reformat.
Does it write my résumé for me?
It proposes edits against one posting, tied to evidence you confirmed, and writes the ones you accept into your own file. It does not compose a résumé out of nothing, and it does not send anything to an employer.
What stops it from inventing things?
Two layers. A proposed bullet must cite evidence ids from your Career Brain, and Truth Guard classifies every edit as verified, reframed or unsupported, and a deterministic lexical re-check sits behind the model and can only lower that verdict, never raise it. An unsupported edit can be read and cannot be accepted. It is a floor, not a guarantee: a claim that borrows the evidence's own words can still get past it, which is why you read the edit before you take it.
Will my formatting survive?
The generated file is your uploaded archive with the accepted replacements written in place; untouched parts are preserved byte for byte. Only edits that anchor exactly to existing text are written — an accepted edit that cannot be anchored is named in the panel rather than silently dropped. PDF output is not built yet.
Do I have to fill in a huge profile first?
No. Import a résumé and confirm the drafts it proposes — that is enough to analyze a posting and tailor against it. Adding rows by hand makes the evidence sharper, but it is not the price of admission.
Does it optimize for the ATS?
No, and it does not claim to. There is no keyword score and no filter-beating mode. What it does is leave your own document intact — the file you download is the archive you uploaded with the accepted edits written in place, not a rebuilt template.
Where does my data go?
Into your own rows, protected by row-level security. When you run a model action, only the fields that action needs are sent to Anthropic — the AI disclosure page lists them task by task. Prompt text, résumé text and model output are not stored in the model log. RoleVerity does not train a model on your data.
Why is there no match score?
Because it would be fiction. You get a verdict and a finding per requirement, each with the evidence that backs it and the gaps it cannot cover — the things you would actually rehearse before an interview.
What is not built yet?
Self-serve checkout, hosted email (so account confirmation only works on a local stack today), PDF export, cover-letter export and error monitoring. The product runs today against a local stack and a founding cohort; it is early, and this page will change as those land.
Who is behind it?
David Shin, an individual operator in California, building it while running a search of his own. hello@roleverity.app reaches him.
Tailor your résumé to the role without adding anything you cannot defend in the interview.
Open the demo on seeded data, or start with your own résumé and one posting you actually want.