AI Job Matcher Agent

The Job Matcher Application is an intelligent workflow designed to automate the job search process. By leveraging LangGraph and GPT-4o-mini, the system analyzes a candidate's CV, performs real-time searches on Google Jobs, and scores listings based on semantic relevance rather than simple keyword matching.

📖 Project Overview

Manual job searching is often inefficient and lacks context. This application solves that by using a State Machine architecture to:

1. Analyze your local vector resume data to infer your ideal job title.
2. Verify the search strategy with you (Human-in-the-Loop).
3. Search live listings via SerpApi.
4. Score every job from 0-100 based on your specific technical skills.

Architecture

The application is built using LangGraph, treating the workflow as a graph of nodes where state is passed and mutated at every step.

graph TD
    Start([Start]) --> Extract[Node: Extract Skills & Title]
    Extract -->|Inferred Title + Skills| Approval{Node: Human Approval}
    
    Approval -->|User Overrides| Search[Node: Search Jobs]
    Approval -->|User Approves| Search
    
    Search -->|Raw Listings| Score[Node: Score Jobs]
    Score -->|Ranked List| Select[Node: Select Top Jobs]
    Select --> End([End])

    style Approval fill:#f9f,stroke:#333,stroke-width:2px
    style Score fill:#bbf,stroke:#333

Loading

Component Breakdown

• Extract Skills Node: Queries a local Chroma database to retrieve CV segments and uses gpt-4o-mini to identify the candidate's core skills and best-fit job title.
• Human Approval Node (HITL): A blocking checkpoint that pauses execution, allowing the user to verify or edit the search query via the console.
• Search Jobs Node: Utilizes SerpApi (Google Jobs engine) to fetch the top 10 live job listings.
• Score Jobs Node: Acts as an "AI Recruiter," assigning a confidence score (0-100) and a reasoning sentence to every job found.

🛠️ Prerequisites & Setup

1. Environment Variables

You must create a .env file in the root directory with the following keys:

OPENAI_API_KEY=sk-...
SERP_API=...

2. Dependencies

The system relies on Python 3.12+ and the following libraries:

• langgraph
• langchain & langchain-openai
• langchain-chroma & langchain-huggingface
• google-search-results (SerpApi)
• python-dotenv

3. Database Configuration

Note: The script currently points to a hardcoded local ChromaDB path. You must update the DB_PATH constant in job_matcher.py to point to your local vector store:

DB_PATH = "/path/to/your/chroma_db"

Usage

Run the main script to initiate the agent workflow:

python job_matcher.py

Runtime Interaction

1. The agent will print the Inferred Job Title and a summary of Extracted Skills.
2. It will pause and wait for your input:

• Press Enter: Accepts the AI's suggestion.
• Type a new title: Overrides the search query.

3. The agent will then search, score, and output the Top 3 Jobs with reasons for why they match your profile.

Project Structure

• job_matcher.py: The main entry point containing the LangGraph definition, node logic, and helper functions.
• requirements_doc.md: Detailed functional and non-functional requirements.
• Tech_Doc.md: Technical walkthrough and code explanation.

⚠️ Notes

• Tokenizers: The script explicitly disables tokenizer parallelism (TOKENIZERS_PARALLELISM="false") to prevent deadlock issues common when using HuggingFace embeddings on macOS.
• Model: The system maps the internal request gpt-4.1-mini to the actual OpenAI model gpt-4o-mini.