a modular AI-powered job application automation system built entirely with n8n

AI Job Hunter

AI Job Hunter is a modular AI-powered job application automation system built entirely with n8n.

The project automates the job-hunting workflow from CV processing and job discovery to semantic job matching, AI-powered CV tailoring, PDF generation, human approval, application tracking, and email confirmation.

Instead of building one large workflow, the system is divided into five independent n8n modules, controlled by a Main Orchestrator.

Features

Automated CV processing and information extraction

Multi-source job discovery

Adzuna and Remotive API integration

Job normalization, filtering, and deduplication

Semantic job matching using embeddings

Hybrid candidate-job relevance scoring

AI-powered CV tailoring

AI output fact-checking

Automatic retry when validation fails

LaTeX CV generation

Automatic PDF compilation and verification

Human-in-the-Loop approval

Duplicate application prevention

Airtable application tracking

Automated email confirmation

Central n8n workflow orchestration

Controlled error handling between modules

Architecture

Candidate CV

|

v

Module 1

CV Processing

|

v

Module 2

Job Search & Filtering

|

v

Module 3

Hybrid Job Matching

|

v

Module 4

AI CV Tailoring & PDF

|

v

Module 5

Application & Tracking

|

v

Final Run Report

Controlled by Main Orchestrator

Workflow Modules

Module 1 — CV Processing & Information Extraction

Objective

Receive the candidate CV, validate the uploaded file, extract readable content, and prepare structured candidate information for the rest of the pipeline.

Workflow

Trigger

-> Normalize Input

-> Needs Download?

-> File Validation

-> Is File Valid?

-> Route by File Type

-> Text Extraction

-> Text Validation

-> AI Extraction

Key Features

Supports manual, workflow, and form triggers

Normalizes incoming CV data

Supports uploaded files and remote file URLs

Validates file type and availability

Routes processing according to file type

Extracts text from PDFs and text-based documents

Checks extracted text quality

Uses Gemini as a PDF fallback when normal extraction is insufficient

Produces structured candidate information

Output

Clean + Validated Candidate Profile

Module 2 — Job Search, Aggregation & Filtering

Objective

Search multiple job sources, normalize their responses, remove duplicate or low-quality listings, and produce a clean list of opportunities.

Job Sources

Adzuna

Remotive

Workflow

Normalize Search Config

|

+--> Build Adzuna Query

| |

| v

| Adzuna API

| |

| v

| Normalize Adzuna

|

+--> Build Remotive Query

|

v

Remotive API

|

v

Normalize Remotive

|

v

Merge Sources

|

v

Flatten Jobs

|

v

Description Mining

|

v

Deduplication

|

v

Quality Filtering

|

v

Assemble Jobs

Because Adzuna and Remotive return different JSON structures, each source is normalized before the results are merged.

Output

Normalized + Deduplicated + Filtered Job List

Module 3 — Hybrid Job Matching

Objective

Compare the candidate profile against discovered job opportunities and calculate relevance scores.

The system goes beyond traditional keyword matching by using semantic embeddings.

Workflow

Normalize Input

|

v

Build Embedding Batch Request

|

v

Get Batch Embeddings

|

v

Compute Hybrid Scores

|

v

Ranked Jobs

Matching Strategy

Candidate Profile

|

v

Embeddings

|

+--------------------+

| |

v v

Semantic Similarity Other Matching Signals

| |

+---------+----------+

|

v

Hybrid Score

|

v

Ranked Jobs

This allows the system to recognize relevant opportunities even when the wording in a CV differs from the wording used in a job description.

Output

Ranked Job Opportunities + Hybrid Scores

Module 4 — AI CV Tailoring & Document Generation

Objective

Tailor the candidate's CV for a selected job while ensuring that the AI does not introduce unsupported information.

Workflow

Normalize Input

|

v

LLM Tailor — Attempt 1

|

v

Parse & Fact-Check

|

v

Fact-Check Passed?

|

+--+--+

| |

YES NO

| |

| v

| Prepare Regeneration

| |

| v

| LLM Tailor — Retry

| |

| v

| Parse & Fact-Check

|

v

Reassemble Tailored LaTeX

|

v

Compile LaTeX to PDF

|

v

Verify Compiled PDF

|

v

Assemble Application Package

|

v

Create PDF Binary

AI Fact-Checking

The system validates AI-generated CV content before generating the final document.

If the first AI response fails validation, the system automatically attempts regeneration.

If the retry still fails, the workflow emits a controlled failure instead of producing an inaccurate CV.

Document Generation

Validated CV Content

|

v

Tailored LaTeX

|

v

PDF Compilation

|

v

PDF Verification

|

v

Final Application Package

Output

The application package contains the candidate information, selected job, tailored CV, generated PDF, and the information required by the next module.

Module 5 — Application & Tracking

Objective

Validate the final application package, prevent duplicate applications, obtain human approval, perform the submission stage, record the application, and send confirmation.

Workflow

Normalize Input

|

v

Validate Package

|

v

Package Valid?

|

v

Duplicate Check — Airtable

|

v

Is Duplicate?

|

v

Wait for Human Approval

|

v

Process Approval Decision

|

v

Approved?

|

v

Mock Submission

|

v

Validate Submission

|

v

Persist Tracking Record

|

v

Prepare Email Attachments

|

v

Send Confirmation Email

|

v

Submitted

Human-in-the-Loop

The system includes a human approval checkpoint before the application submission stage.

Application Ready

|

v

Wait for Human Approval

|

+---+---+

| |

APPROVE REJECT

| |

v v

Continue Stop

This keeps the candidate in control of the final application decision.

Duplicate Application Protection

Before continuing with an application, the system searches Airtable using the candidate and job identifiers.

This prevents repeated applications to the same opportunity.

Application Tracking

Successful application information is stored in Airtable, providing a persistent record for application tracking and future duplicate detection.

Email Confirmation

After the submission and tracking stages are completed, the workflow prepares the relevant application attachments and sends an automated confirmation email to the candidate.

Current Submission Status: The submission stage currently uses a Mock Submission for development and testing. Real ATS or supported job-platform integrations can be added as future extensions.

Main Orchestrator

The Main Orchestrator controls the complete n8n system.

Rather than placing every node inside one massive workflow, the application is separated into independent modules.

Orchestrator Workflow

Normalize Orchestrator Input

|

v

Execute Module 1

|

v

Succeeded?

|

v

Execute Module 2

|

v

Succeeded?

|

v

Execute Module 3

|

v

Succeeded?

|

v

Execute Module 4

|

v

Succeeded?

|

v

Execute Module 5

|

v

Final Run Report

Every module is checked before the next module is allowed to execute.

If a module fails, the orchestrator stops the pipeline safely and reports the failed stage.

Complete End-to-End Flow

Candidate Uploads CV

|

v

CV Processing

|

v

Candidate Extraction

|

v

Job Search

|

v

Job Source Normalization

|

v

Deduplication

|

v

Quality Filtering

|

v

Semantic Embeddings

|

v

Hybrid Job Matching

|

v

Job Ranking

|

v

CV Tailoring

|

v

Fact Checking

|

v

PDF Generation

|

v

Package Validation

|

v

Duplicate Check

|

v

Human Approval

|

v

Submission Stage

|

v

Airtable Tracking

|

v

Email Confirmation

|

v

Final Run Report

Technology Stack

Technology Role

n8n Core automation platform, workflow logic, and orchestration

AI / LLM APIs CV extraction and tailoring

Embeddings Semantic candidate-job matching

Gemini PDF extraction fallback

Adzuna API Job discovery

Remotive API Remote job discovery

Airtable Application tracking and duplicate detection

LaTeX Tailored CV document generation

PDF Compilation Service Final PDF generation

Gmail / Email Candidate confirmation

Repository Structure

AI-Job-Hunter/

|

├── workflows/

│ ├── 01-CV-Processing.json

│ ├── 02-Job-Search.json

│ ├── 03-Hybrid-Job-Matching.json

│ ├── 04-AI-CV-Tailoring.json

│ ├── 05-Application-Tracking.json

│ └── Main-Orchestrator.json

|

├── screenshots/

│ ├── module-1.png

│ ├── module-2.png

│ ├── module-3.png

│ ├── module-4.png

│ ├── module-5.png

│ └── main-orchestrator.png

|

├── docs/

│ └── technical-report.md

|

├── .gitignore

├── LICENSE

└── README.md

Setup

1. Access n8n

The project is built entirely using n8n workflows.

You can use:

n8n Cloud

A self-hosted n8n instance

2. Clone the Repository

git clone https://github.com/YOUR_U...

cd AI-Job-Hunter

3. Import the Workflows

Open your n8n instance and import the workflow JSON files from the /workflows directory.

Import the five modules and the Main Orchestrator.

4. Configure Credentials

Configure the required credentials directly inside your n8n instance.

Depending on your configuration, these may include:

AI / LLM Provider

Embedding Provider

Gemini

Adzuna

Airtable

Gmail / Email

PDF Compilation Service

5. Configure the Main Orchestrator

After importing the workflows, verify that each Execute Workflow node inside the Main Orchestrator points to the correct imported module.

Then configure the required input parameters and execute the orchestrator.

Security

Never commit API keys, access tokens, passwords, or credentials to GitHub.

n8n credentials should be configured through the n8n credential system.

Before publishing exported workflow JSON files, inspect nodes such as:

HTTP Request

Code

Set / Edit Fields

Headers

Query Parameters

Webhook URLs

Make sure no private API key, token, email address, personal CV data, or other sensitive information has been hard-coded into the workflow.

Design Principles

Modular Architecture

Each workflow performs one clear responsibility, making the project easier to maintain, test, debug, extend, and reuse.

Human Control

AI assists with the application process, but a human approval checkpoint remains before submission.

Semantic Matching

Embeddings provide deeper candidate-job matching than simple keyword comparison.

AI Validation

AI-generated CV content is fact-checked before document generation.

Controlled Failure

Validation and success gates prevent invalid data from propagating through the pipeline.

Fallback Processing

Gemini provides an alternative extraction path when normal PDF extraction is insufficient.

Application Deduplication

Airtable tracking prevents repeated applications to the same job.

Current Project Status

Feature Status

CV Processing Complete

CV Information Extraction Complete

Adzuna Job Search Complete

Remotive Job Search Complete

Job Normalization Complete

Deduplication Complete

Quality Filtering Complete

Semantic Embeddings Complete

Hybrid Matching Complete

AI CV Tailoring Complete

AI Fact Checking Complete

Automatic AI Retry Complete

LaTeX Generation Complete

PDF Compilation Complete

PDF Verification Complete

Human Approval Complete

Airtable Tracking Complete

Duplicate Detection Complete

Email Confirmation Complete

Main Orchestrator Complete

Real Job Submission Mock / Future Integration

Future Improvements

Potential future improvements include:

Real ATS integrations

Additional job-search APIs

LinkedIn job discovery where supported

Automatic cover-letter generation

Advanced salary filtering

Location preference scoring

Candidate preference learning

Application analytics

Interview tracking

Recruiter follow-up workflows

Additional database integrations

Vector database support

Candidate dashboard

Project Goal

AI Job Hunter is designed as a controlled, intelligent, and modular job-hunting automation pipeline.

It combines:

n8n Automation

+

Job Discovery

+

Semantic Matching

+

Generative AI

+

Fact Checking

+

Human Approval

+

Application Tracking

The goal is to make the job application process more targeted, traceable, and reliable while maintaining human control over the final submission decision.