Relevant Job Finder 🇵🇰

A RAG (Retrieval-Augmented Generation) based job recommendation system for the Pakistani job market. It fetches fresh job postings from the web, builds a semantic search index, and uses a cloud LLM to recommend the most relevant jobs based on your query — in plain English.

---

How It Works

Search Keyword → JSearch API → Fresh Jobs CSV
                                      ↓
                              Data Loader (LangChain Documents)
                                      ↓
                         HuggingFace Embeddings + ChromaDB (in-memory)
                                      ↓
                     MMR Retrieval → Deduplication → Custom Retriever
                                      ↓
                        User Query → RAG Pipeline → Groq LLM
                                      ↓
                              Job Recommendations (Streamlit UI)

1. CSV Maker — fetches live job postings from JSearch API based on a search keyword, uses Groq LLM to extract structured fields from job descriptions, and saves them as a CSV
2. Data Loader — reads the CSV, cleans it, and converts each job into a LangChain Document object
3. Vectorizer — embeds all job documents using HuggingFace (all-MiniLM-L6-v2) and stores them in an in-memory ChromaDB vectorstore
4. Retriever — uses MMR (Maximal Marginal Relevance) search to fetch diverse relevant documents, deduplicates them, and wraps them in a custom StaticListRetriever
5. Responder — passes retrieved jobs to Groq LLM via a custom prompt and generates structured job recommendations
6. Streamlit UI — a clean web interface where users enter their job preferences and get recommendations instantly

---

Project Structure

Relevant-Job-Finder/
│
├── src/
│   ├── data/
│   │   ├── csv_maker.py        # Fetches jobs from JSearch API → saves CSV
│   │   └── data_loader.py      # Loads CSV → LangChain Document objects
│   │
│   ├── features/
│   │   ├── vectorizer.py       # Embeds documents → in-memory ChromaDB
│   │   └── retriever.py        # MMR search → deduplication → StaticListRetriever
│   │
│   └── generation/
│       └── responder.py        # RAG chain → Groq LLM → job recommendations
│
├── dataset/                    # Auto-created, stores fetched jobs CSV
├── app.py                      # Streamlit entry point
├── config.json                 # All configuration in one place
└── requirements.txt

---

Setup & Installation

1. Clone the repository

git clone https://github.com/yourusername/relevant-job-finder.git
cd relevant-job-finder

2. Create and activate virtual environment

python -m venv .venv
.venv\Scripts\activate        # Windows
source .venv/bin/activate     # Mac/Linux

3. Install dependencies

pip install -r requirements.txt

4. Get a JSearch API key

Sign up for free at rapidapi.com and subscribe to the JSearch API (free tier). Copy your API key.

5. Get a Groq API key

Sign up for free at console.groq.com and generate an API key.

6. Configure config.json

{
    "groq_model_csvmaker": "llama-3.1-8b-instant",
    "groq_model_responder": "llama-3.3-70b-versatile",
    "search_keyword": "AI/ML Internships",
    "csv_path": "dataset",
    "csv_name": "fetchedjobs.csv",
    "jsearch_api": "<your JSearch API key here>",
    "Huggingface_embedder_model": "all-MiniLM-L6-v2",
    "groq_api_key": "<your Groq API key here>"
}

---

Usage

Run the Streamlit app:

streamlit run app.py

Then:

1. Enter your job preferences in the text box (e.g. "I am looking for remote AI internships")
2. Click Find Jobs
3. The system fetches fresh jobs, builds a semantic index, and returns personalized recommendations with apply links

---

Configuration Options

Key	Description
groq_model_csvmaker	Groq model used to extract job fields from descriptions
groq_model_responder	Groq model used to generate job recommendations
search_keyword	Keyword sent to JSearch API to fetch jobs
csv_path	Folder where the fetched jobs CSV is saved
csv_name	Name of the generated CSV file
jsearch_api	Your RapidAPI key for JSearch
Huggingface_embedder_model	HuggingFace sentence-transformer model for embeddings
groq_api_key	Your Groq API key for LLM inference

---

Tech Stack

Component	Technology
Job Data	JSearch API (via RapidAPI)
Embeddings	HuggingFace all-MiniLM-L6-v2
Vector Store	ChromaDB (in-memory)
Retrieval	MMR Search + Custom StaticListRetriever
LLM	Groq (llama-3.1-8b-instant, llama-3.3-70b-versatile)
RAG Framework	LangChain
Data Processing	Pandas
Frontend	Streamlit

---

Requirements

langchain
langchain-community
langchain-classic
langchain-ollama
langchain-huggingface
langchain-core
langchain-groq
chromadb
sentence-transformers
ollama
requests
pandas
streamlit

---

Notes

• The vectorstore is in-memory only — it rebuilds fresh on every run with the latest job data
• The dataset/ folder is auto-created if it doesn't exist
• JSearch free tier allows up to 200 requests/month — more than enough for personal use
• Groq free tier is sufficient for personal and portfolio use
• Add config.json to .gitignore to avoid exposing your API keys — use config.example.json as a template instead