Pursuiter: AI-assisted job matching platform
Mohammad Al-Qasem
Pursuiter
A two-sided AI job board that flips the signal-to-noise ratio: applicants get pre-application feedback, recruiters get fewer-but-better-qualified applications.
Overview
Pursuiter is a job-application platform with a twist on both sides:
For applicants — paste your résumé and a job listing, and a Gemini-powered analyzer grades fit, surfaces missing skills, and suggests phrasing tweaks before you submit.
For recruiters — set explicit minimum criteria (skills, years, education) and only see applications that meet the bar, with the AI's notes attached.
The point: cut the wasted time on both sides. Applicants stop blasting unfit roles; recruiters stop sifting through unqualified resumes.
Why this exists
Built end-to-end as a personal full-stack project to learn the MERN stack with strict architectural separation (MVC) and to integrate an LLM (Gemini) behind a swappable adapter rather than calling the SDK from controllers.
Stack
Layer Choice Frontend React 18 Backend Node.js + Express Database MongoDB (via Mongoose) Auth JWT AI Google Gemini API (behind an adapter) Tests Jest + Supertest
Architecture
Strict three-layer Model-View-Controller:
Model — Mongoose models in
/backend/models/ define the shape and validators. No business logic.View — React components in
/frontend/src/components/ are presentation only. No data fetching or business logic.Controller — Express controllers in
/backend/controllers/ hold all business logic. Each controller is responsible for one resource (Users, Jobs, Applications, AI).The Gemini integration sits behind a thin adapter (
/backend/ai/) so the model provider can be swapped without touching controller code or the React app.Key engineering decisions
MVC separation as a hard rule. No business logic in React; no presentation logic in controllers. This made the test surface easy to define — every business rule has a controller test, not a UI test.
AI behind an adapter. The Gemini call lives in one place. Want to A/B test against another model? Replace one file. The controller and the UI never know.
Recruiter-set criteria as filters, not blockers. Applicants still see jobs they don't quite fit (so they can grow into them) — the AI just tells them so up front. Only submission requires meeting the bar.
JWT auth with httpOnly refresh tokens. Standard separation; refresh tokens stored httpOnly to mitigate XSS exfiltration.
Setup
Prerequisites
Node.js 22+
MongoDB 7+
Google Gemini API key
Configure
Create
.env files in /backend and /frontend:Run
Test
Screenshots
Landing
Applicant dashboard
Recruiter dashboard
Recruiter applicant view
Project structure
Contribution
Branching strategy:
dev is the integration branch; feature branches branch from dev, named by ticket number. PRs go to dev, require one peer review, and dev merges to main at sprint boundaries.Code style
React components live under
/frontend/src/components/<kebab-case>/ with a CamelCase JS + matching CSS file (e.g. ApplicantDashboard.js + ApplicantDashboard.css).Express controllers live under
/backend/controllers/ with CamelCase filenames (e.g. UserController.js).No business logic in React components; no presentation in controllers.
License
Personal portfolio project.
Like this project
Posted Aug 8, 2026
Built a two-sided MERN job board where Gemini gives applicants fit feedback before submission and recruiters review candidates against defined criteria.
