Inside the four quiet steps.
We get asked, often, what the model is actually doing. Here is the whole flow — every pipeline, every guardrail, in plain language.
Four quiet steps
between your résumé and the role.
No dashboards to learn. No prompts to write. Just the answers you needed.
Upload once.
We parse your résumé into a structured profile and reuse it for every analysis. PDF or DOCX, up to 5 MB.
Paste any job.
Drop a URL or paste the description. We read the requirements and weigh them against your profile.
Read the verdict.
A match score, a skill gap map, and a click on any skill explains how to close it.
Generate. Apply.
Tailored CV, cover letter, and an interview prep pack — verified for no fabrication.
Reading your résumé without flattening it.
When you upload a PDF or DOCX, our Strong AI extracts the document into a structured profile — your contact details, every role with dates and accomplishments, and the skills the work implies. The result is validated against a strict schema, so what we keep is exactly what we expect: name, contact, education, experience entries, and a deduplicated skill set. Your résumé text is never logged — only an opaque event with your user id and the cost incurred.
Reading the job like a tired hiring manager.
Paste the description or drop a URL. We extract the requirements that actually matter — the must-haves, the nice-to-haves, the years and the seniority — and weigh each one against your profile. If anything is fuzzy, we err toward listing it as a requirement so the score reflects the real gap, not a wishful one.
The verdict, with reasoning you can click into.
Our Powerful AI makes the judgment call: a single match score and a paragraph of narrative that explains why. Every skill in the gap map is clickable — that triggers explainSkill, a cached pipeline that returns one short, specific paragraph on how to close the gap. We cache by skill and profile, so a second click costs nothing.
Tailoring without inventing.
We rewrite your CV against the job, draft a cover letter, and generate an interview prep pack. Then a structural verifier called noFabCheck reads the output back and compares every claim against your original profile. If it spots a fabricated company, role, date, or skill, the pipeline retries with the violation in the next prompt — up to three times. If we still can't produce a clean draft, the request fails visibly. We will never quietly ship a CV that lies on your behalf.
Two languages, the same care.
EN and AR each have their own prompt files (prompts/cv.en.ts, prompts/cv.ar.ts, and so on) — Arabic isn't a translation of English, it's a first-class voice. UI renders in Cairo with proper RTL handling. Generated PDFs and DOCX use NotoNaskhArabic, with bidirectional handling correct down to the punctuation, so your Arabic résumé doesn't look like an afterthought.