Skip to content
FLOWer Recruit

Case study

From a blank page to a CV made for one offer

FLOWer Recruit is a job-search tool for people who are willing to put in some effort: one honest base CV, then a version for each offer, without claims they could not defend in an interview.

The candidate’s problem

A CV pulled from a drive is usually a list of generalities. A chat model will happily “improve” it, and in doing so add things the candidate cannot back up in an interview.

Every offer needs a different emphasis. Copying and hand-editing several versions drifts apart quickly, and nobody remembers which one says what.

Pasting an offer into a chat gives an analysis anyone can get in a minute. It does not give a CV built on facts that were checked.

The path to the first CV

  1. Language

    The interface language comes first, before any document is written.

  2. Base CV with CV Architect

    A conversation draws out concrete achievements. A file or a LinkedIn profile can be raw material, but the base CV is assembled only from facts the candidate confirmed.

  3. Job offer

    The offer is pasted or imported with the Chrome extension and becomes its own recruitment process.

  4. ATS Match & CV Builder

    Requirements are compared with the base CV. Gaps turn into questions to the candidate; the tailored CV is written from the base CV and confirmed answers only.

  5. Download

    The version for this offer is saved with its process and exported as PDF or DOCX.

Decisions you can see in the product

Facts are confirmed, not inferred

The base CV is the only source CV Builder writes from. When an offer asks for something the base CV does not show, the candidate gets one direct question. A bare “yes” changes nothing; a sentence about what they did and where does.

Base CV and offer versions are separate

The base CV lives in the CV library. Every tailored version belongs to one recruitment process, with its own history, so tailoring for one offer never rewrites the reference.

A file is material, not a finished CV

A shorter onboarding was tried: offer first, with an uploaded file saved as the base CV unchecked. It produced fast analyses of an unverified CV and was reverted to language, then base CV, then offer.

The score is computed, not generated

The model is not allowed to state a score. The server calculates it from the final status of each requirement, so the number follows the evidence.

A light public entry

Public pages are rendered to HTML at build time and do not download the application until someone signs in. A strict Content-Security-Policy allows only hashed inline scripts.

Screens

CV Architect conversation with a fictional candidate.
CV Architect records the facts the candidate confirmed.
Original job description of a fictional offer.
The offer is stored with its own recruitment process.
ATS Match requirements for a fictional offer.
ATS Match marks covered requirements and honest gaps.
Download dialog for the tailored CV.
The tailored CV downloads as PDF or Word in five styles.

Screens show a fictional candidate and a fictional offer.

How it is built

  • Browser: React, Vite and Zustand, with the authenticated app loaded on demand.
  • Server: Express with Prisma on PostgreSQL; shared TypeScript contracts used by both sides.
  • Hosting: Vercel for the frontend, Railway for the API and Neon for the database.

Back to FLOWer Recruit