How it works
Four stages, one constraint.
The constraint is that nothing reaches your resume unless you put it in your master profile first. Every stage below is shaped by it.
The master profile
Written once, by you. This is the only place resume content can come from.
You write your achievements as complete, defensible lines — the way they should read on the page. Each one is stored with a stable identifier and grouped by role.
Alongside them sit the rules that are specific to you: which of your employer’s client names may never appear, which figure is canonical when you have described the same accomplishment two different ways, and which skills you can genuinely defend under questioning versus which you have only touched.
Some fields are locked outright — your name, contact details, employers, job titles, and date ranges. These are byte-identical on every document Lapidly ever produces, because they are the fields that get verified against employment records.
The match report
Before anything is built, an honest read on whether to apply at all.
Paste a posting. Lapidly maps every requirement to your evidence and marks it strong, partial, or absent, citing the specific bullet that supports each one. A requirement with nothing behind it is reported as a gap rather than quietly papered over.
Coverage is stated as a count — meets 7 of 7 required — never as a percentage. A count is checkable line by line against the posting. A percentage is a number with a confident shape and no method behind it.
Then it stops and waits. Building a resume for every posting you find is how you end up with twenty-one applications and no interviews. A gate that sometimes says skip this one is doing its job.
- Strong
5+ years enterprise technical support
8 years across Checkr and Salesforce
- Strong
OAuth 2.0, SAML, SCIM troubleshooting
Subject matter expert; 40+ identity issues monthly
- Strong
SQL and log analysis
Snowflake dashboards, Datadog, Kibana
- Absent
Kubernetes in production
Fundamentals only — reported as a gap
Meets 3 of 4 required. The gap is real and worth naming in the interview rather than hiding.
Selection
The model chooses. It does not write.
Given the posting and your master profile, the model returns which bullets to use, in what order, which summary components to combine, and how to order your skills so the posting’s stack leads.
What it returns is a list of identifiers. The assembler looks each one up in your master profile and refuses anything that does not resolve. A sentence the model composed has no route onto the page — not because it was instructed not to compose one, but because the format of its answer cannot carry one.
Selection and ordering are where tailoring genuinely happens. Leading a role with the bullet that answers the posting’s first requirement matters more than any rewording would.
The gate
The document is read back before you can send it.
The assembled document runs through the deterministic guards — customer names, tier language, unverified skills, locked fields, metric canon, and a provenance check confirming every bullet matches your master profile exactly.
Then it is rendered and read back by two independent parsers. Most text-extraction libraries wrap the same underlying engine, so pairing two of them checks that engine against itself and always agrees, including when it is wrong. Lapidly’s second engine reads the PDF’s own drawing instructions directly and shares no code with the first.
If they disagree about a single word, that is a failure. The system on the other end is a third parser nobody can inspect, and a document that two engines already read differently is a coin flip.
A blocked document is not deliverable. There is no override, because an override is how a fail-closed system becomes a fail-open one on the first busy afternoon.
What you get with each build
A tailored DOCX and PDF, plus two things most tools do not give you: a coverage table separating what is genuinely evidenced from what is thin, and a stretch list naming the three or four bullets pushed hardest — so nothing on your own resume can surprise you in an interview.
Submit the DOCX wherever it is accepted. Greenhouse, Lever, and Ashby all take it, and it parses more reliably than PDF everywhere, because the structure is explicit rather than inferred from where the glyphs happen to sit.