ApplyEase — Local, Privacy‑First Job Application Assistant

Overview

• Generates Tailored CV based on Job Application.

• Auto‑fills common job application fields (first/last name, email, phone, links) and uploads your resume.

• Computes a Resume ↔ Job Description match score with tech‑term highlights (matching/missing keywords).

• Generates concise custom answers to application questions using a local LLM (no paid APIs).

• NEW: Structured resume builder (summary, title, skills, multiple experiences, education) with templated tailored CV generation and PDF export.

• NEW: Cover letters — generate LLM‑enhanced letters or clean templates; saved history with downloads.

• NEW: Job Tracker — board and list view, drag‑and‑drop between stages (Saved → Applied → Interview → Offer → Rejected), quick cover‑letter button.

• NEW: Extension auto‑tracking — captures an “Applied” job automatically on submit for many job sites.

• Works via a Chrome extension with a small React dashboard and a FastAPI backend.

Architecture

• Backend: FastAPI (applyease-backend/app.py) + PostgreSQL with pgvector for embeddings. SentenceTransformer all-MiniLM-L6-v2. Routers: routes/job_tracker.py, routes/cover_letters.py. Local LLM: Ollama (default) or LM Studio/OpenAI‑compatible.
• Frontend: React app in frontend/ (login, dashboard, resume builder, cover letters tab, job tracker board/list).
• Chrome Extension: Autofill, JD extraction, on‑page match widget, popup with keywords, job tracker button, auto‑tracking.

Prerequisites

• Python 3.9+
• PostgreSQL with pgvector extension available
• Node.js 16+ and npm for the frontend
• Chrome (or Chromium‑based) for the extension
• Local LLM
  • Ollama (recommended): https://ollama.ai — e.g., ollama pull llama3.1:8b
  • OR LM Studio / any OpenAI‑compatible local server

Quick Start

1. Database

   • Create DB applyease and ensure pgvector is installed/enabled
     • psql: CREATE DATABASE applyease; \c applyease; CREATE EXTENSION IF NOT EXISTS vector;

2. Backend

   • cd applyease-backend
   • python3 -m venv .venv && source .venv/bin/activate
   • pip install -r requirements.txt
   • Optional env (defaults in parentheses):
     • DB: PGHOST (localhost), PGPORT (5432), PGUSER (your OS user or postgres), PGDATABASE (applyease), PGPASSWORD (empty)
     • Auth: JWT_KEY (dev-secret), JWT_EXPIRES_IN_MIN (60)
     • LLM: LLM_PROVIDER (ollama|lmstudio), LLM_MODEL (e.g., llama3.1:8b), OLLAMA_HOST (http://localhost:11434), LLM_BASE_URL (http://localhost:1234/v1)
   • Start: uvicorn app:app --reload --port 8000
   • On first run, the service will create tables users, resumes and try to create the vector extension and an IVFFlat index.

3. Local LLM

   • Ollama (recommended):
     • Install Ollama and run: ollama pull llama3.1:8b
     • export LLM_PROVIDER=ollama and export LLM_MODEL=llama3.1:8b
   • LM Studio (or compatible):
     • Start local server (usually http://localhost:1234/v1) with your model
     • export LLM_PROVIDER=lmstudio, export LLM_BASE_URL=http://localhost:1234/v1, export LLM_MODEL=<your-model-name>

4. Frontend

   • cd frontend
   • npm install
   • Optional: export REACT_APP_API_BASE=http://localhost:8000
   • Run: npm start (dev server on http://localhost:3000)

5. Chrome Extension

   • Go to chrome://extensions → Enable Developer Mode → Load unpacked → select this repo folder.
   • The extension injects on all frames at document_end for better compatibility.

Basic Flow

1. Sign up or login in the web app (http://localhost:3000). JWT is saved to localStorage and also broadcast to the extension.
2. In the dashboard, upload your resume (PDF) and update profile fields. This embeds and stores the resume in Postgres.
3. Visit a job posting (LinkedIn/Indeed/Workday/Greenhouse/Lever/etc.).
   • A floating “Resume Match: XX%” widget appears on the page (computed from the JD on load).
   • Click the widget to open the extension popup, which shows matching/missing keywords.
4. Click “Auto Fill” in the popup to fill the form.
   • Fills first/last name (or full name field), email, phone, location, URLs.
   • Attaches your resume as a PDF from the backend (/resume_pdf).
   • Adds “Fill” buttons next to textareas to generate tailored answers locally.
5. Resume Builder and Tailoring
   • Build your resume sections (title, summary, skills, experiences with bullets, education) and save.
   • Paste a JD and click “Generate Tailored CV” to render a templated PDF that prioritizes relevant experiences.
   • Click “Generate Cover Letter” to create a letter PDF using the JD and your profile/sections (uses local LLM if configured).
6. Job Tracker
   • Open Job Tracker from the dashboard or extension popup.
   • Add jobs or drag cards between columns to update status; use List view if preferred.
   • Auto‑tracking: on many sites, after you submit an application, the extension records the job with status “applied”.

Key Endpoints (Backend on :8000)

• POST /signup → { token, user }

• POST /login → { token }

• GET /user (Bearer)

• PATCH /user (Bearer, multipart: resume file + fields first_name,last_name,urls JSON, etc.)

• GET /resume (Bearer) → { resume_text }

• GET /resume_pdf (Bearer) → stream of generated PDF

• POST /match (Bearer) → { percent, matchingWords, missingWords, score }

• POST /custom-answer (Bearer, local LLM only) → { answer }

• POST /upsert_resume (service) → { ok, user_id }

• Structured resume sections

  • GET /resume_sections (Bearer) → { summary, title, experiences, education, skills }
  • POST /resume_sections (Bearer) → upserts sections and refreshes embedding/keywords

• Tailored resume (templated)

  • POST /generate_tailored_resume (Bearer) → { id, download_url } (PDF saved in history)
  • GET /tailored_resumes / GET /tailored_resume_download?id=...

• Cover letters

  • POST /cover_letters/generate (Bearer) → PDF stream; body supports { company, title, job_description, filename, save, use_llm }
  • GET /cover_letters, GET /cover_letters/download?id=...

• Job tracker

  • GET /jobs (Bearer) → array of jobs
  • POST /jobs (Bearer) → create job { company, title, ... }
  • PATCH /jobs/{id} (Bearer) → partial update (e.g., { status: 'interview' })
  • DELETE /jobs/{id} (Bearer)
  • GET /jobs/stats (Bearer)

Database Schema (auto‑created)

• users(id text pk, first_name, last_name, email unique, password_hash, phone, location, urls jsonb, eeo jsonb, created_at, updated_at)
• resumes(user_id text pk, resume_text text, embedding vector(384), resume_keywords text[], updated_at)
  • Plus structured fields: summary text, title text, experiences jsonb, education jsonb, skills text[], and optional stored resume file (blob/mime/filename)
• tailored_resumes(id text pk, user_id, job_description, resume_text, resume_blob, resume_mime, resume_filename, created_at)
• cover_letters(id text pk, user_id, job_id, company, title, letter_text, letter_blob, letter_mime, filename, created_at)
• job_applications(id text pk, user_id, company, title, location, source, url, status, notes, jd_text, next_action_date, created_at, updated_at)

Smoke‑Test (curl)

• Signup: curl -sS -X POST http://localhost:8000/signup -H 'Content-Type: application/json' -d '{"first_name":"Test","last_name":"User","email":"test@example.com","password":"pass123"}'
• Upsert resume (text): curl -sS -X POST http://localhost:8000/upsert_resume -H 'Content-Type: application/json' -d '{"user_id":"<USER_ID>","resume_text":"Software engineer skilled in Python, AWS, and Docker."}'
• Health: curl -sS http://localhost:8000/healthz
• Match (Bearer): curl -sS -X POST http://localhost:8000/match -H "Authorization: Bearer <TOKEN>" -H 'Content-Type: application/json' -d '{"jobDescription":"Looking for backend developer with AWS and Python."}'
• Custom answer (Bearer, local LLM): curl -sS -X POST http://localhost:8000/custom-answer -H "Authorization: Bearer <TOKEN>" -H 'Content-Type: application/json' -d '{"jobDescription":"...","applicationQuestion":"Describe a project that demonstrates your impact."}'

Troubleshooting

• Postgres/pgvector: If startup fails creating the vector extension, install pgvector on your DB and retry. Ensure env PG* vars point to the right DB.
• Autocommit error: We set autocommit before using pgvector; ensure you’re on the latest applyease-backend/app.py.
• Local LLM:
  • Ollama: ensure the model is pulled and LLM_MODEL is set (e.g., llama3.1:8b).
  • LM Studio: verify base URL /v1 and the model name; set envs as above.
• CORS: The backend enables permissive CORS so the web app and content scripts can call it. Restart the server after installing deps.
• Extension not autofilling:
  • Confirm token: open DevTools on the job page and run chrome.storage.local.get('token', console.log).
  • Some sites load forms in iframes; we inject into all frames (all_frames: true).
  • Resume upload: Browsers block programmatic file pickers. We set input.files via DataTransfer when allowed; otherwise we attempt drag‑and‑drop onto known dropzones.
  • Auto‑tracking: If a job isn’t captured, it likely didn’t expose title/company reliably; we can add site‑specific selectors upon request.
• Matching seems low with all terms: The score is cosine similarity of embeddings, not a keyword count. Keywords are diagnostic only.

Notes

• Embeddings model: all-MiniLM-L6-v2 (384‑dim). Vectors stored normalized; cosine via dot product.
• Keywords: tech‑focused extraction with curated allowlist + heuristics; generic job terms ignored.
• Data lives in Postgres; to reset, drop the resumes and users tables.

License

• For personal use. Do not upload sensitive information to third‑party sites without review.