JobFinder

Motivation

I wanted to ship a credible end-to-end product: real job data, auth, saved state, and AI that actually helps decisions instead of only generating blurbs. Job boards are noisy; the interesting problem is helping someone narrow roles, compare a shortlist, and connect their profile to what employers ask for.

Overview

JobFinder is an AI-assisted job search app on Next.js. Users browse listings from the Adzuna API across many countries, search and filter, save jobs, and compare two or three roles side by side with Gemini-powered summaries. Profile and skills support resume parsing (PDF and DOCX) and AI skill suggestions. An admin dashboard manages users, listings, and AI summaries.

Core product ideas

• Browse and discovery with filters, trending, and recommended sections so the feed stays useful, not endless.
• Compare jobs: pick two or three listings and get a structured comparison (similarities, differences, recommendation, match scores).
• Profile and skills tied to the JD: summaries, optional skill-to-JD match, and resume parsing to reduce manual data entry.
• Admin surface for users, listings, and summaries plus streamed dashboard insights for operations.

Technical approach

I used a pnpm monorepo so shared Zod schemas and UI could live in packages and stay aligned with Mongoose models on the server. That single source of truth cut drift between API contracts and forms. Search and AI outputs are cached in MongoDB with configurable TTL so repeat queries stay fast without hammering Adzuna or Gemini. Auth uses short-lived access tokens and refresh cookies; admin routes are explicitly guarded.

Tech stack

• Monorepo with pnpm workspaces
• Frontend: Next.js (App Router), React, TypeScript, TanStack Query, Tailwind CSS, shadcn/ui, framer-motion
• Backend: REST API, MongoDB with Mongoose, Zod for validation
• AI via Google Gemini (summaries, resume parsing, skill suggestions)
• Job data from the Adzuna API
• Accessibility and responsive layout prioritized

What I learned

• Aligning filter and sort UX with how people actually search matters more than adding more data on the page.
• Shared schemas between client and server pay off quickly on a full-stack app with many endpoints.
• Caching AI and search results is a product decision: freshness vs cost and latency needs explicit defaults and env tuning.
• Shipping compare and profile flows end-to-end taught me to keep key actions obvious and to validate admin-only paths as carefully as user paths.

The full codebase is on GitHub if you want to trace API design, models, or UI patterns.