🤖 HR Policy Assistant

🚀 Live Demo: HR Policy Assistant

🌐 Deployment Platform: Render

A lightweight Retrieval-Augmented Generation (RAG) application for answering employee questions from an internal HR policy document. The project combines a local policy corpus, embeddings, semantic retrieval, and an LLM-powered agent to provide grounded answers instead of relying on memory alone.

🎯 Why this project?

Organizations often store policies in long documents such as employee handbooks, onboarding guides, and internal policy PDFs. Searching these documents manually is time-consuming and error-prone. This project demonstrates how to build a simple but effective HR assistant that:

• reads and splits a policy document into searchable chunks
• converts chunks into vector embeddings
• retrieves the most relevant policy passages
• uses a language model to answer based only on retrieved evidence
• exposes the assistant through a CLI or a Streamlit web app

✨ Features

• HR policy Q&A using RAG
• FAISS vector search for fast retrieval
• Jina embeddings for semantic matching
• Groq-hosted LLM support
• LangChain agent workflow with tool-based retrieval
• Streamlit UI for interactive conversations
• Persistent FAISS index to avoid reindexing on every run
• Easy setup using environment variables and a Python virtual environment

🛠️ Tech Stack

• Python 3.11
• LangChain and LangChain Community
• Groq LLM provider
• Jina Embeddings
• FAISS for vector storage and retrieval
• Streamlit for the web interface
• Python-dotenv for environment configuration
• FAISS local index stored in the project data directory

🏗️ Architecture

The project follows a standard RAG pattern:

1. Document ingestion

   • The HR policy text file is loaded from the data folder.

2. Chunking

   • The document is divided into smaller text chunks with overlap to retain nearby context.

3. Embedding generation

   • Each chunk is converted into vectors using Jina embeddings.

4. Vector storage

   • The embeddings are stored in a FAISS index for fast similarity search.

5. Retrieval

   • A retriever fetches the top matching chunks for the user question.

6. Agent reasoning

   • A LangChain agent passes the retrieved context to an LLM and answers using the search results as evidence.

7. User interface

   • The assistant can be used through the terminal demo or a Streamlit chat app.

A simple view of the flow looks like this:

HR policy text file
        ↓
Document loader
        ↓
Text splitter
        ↓
Embedding model
        ↓
FAISS vector store
        ↓
Retriever
        ↓
LangChain agent + LLM
        ↓
Answer to the user

📁 Project Structure

HR-POLICY-ASSISTANT/
├── app.py                     # Streamlit web app
├── main.py                    # CLI demo entry point
├── requirements.txt           # Python dependencies
├── README.md                  # Project documentation
├── LICENSE                    # MIT license
├── data/
│   ├── hr_policy.txt          # Source HR policy content
│   └── faiss_index/           # Saved FAISS index files
└── hr_assistant/
    ├── __init__.py
    ├── agent.py               # LangChain agent setup
    ├── config.py              # App configuration and environment variables
    ├── document_loader.py     # Loads raw policy text
    ├── embeddings.py          # Jina embedding model wrapper
    ├── llm.py                 # LLM setup
    ├── pipeline.py            # Main orchestration for the assistant
    ├── splitter.py            # Chunking logic
    ├── tools.py               # Retrieval tool used by the agent
    └── vector_store.py        # FAISS build/save/load logic

🚀 Getting Started

📋 Prerequisites

• Python 3.11+
• A Groq API key
• A Jina API key
• Git

1️⃣ Clone the repository

git clone https://github.com/jaswitha112/HR-POLICY-ASSISTANT.git
cd HR-POLICY-ASSISTANT

2️⃣ Create a virtual environment

python -m venv venv

On Windows:

venv\Scripts\activate

On macOS/Linux:

source venv/bin/activate

3️⃣ Install dependencies

pip install -r requirements.txt

4️⃣ Configure environment variables

Create a .env file in the project root with the following values:

GROQ_API_KEY=your_groq_api_key
JINA_API_KEY=your_jina_api_key

The app loads these automatically from the environment using python-dotenv.

5️⃣ Run the project

💻 CLI demo

python main.py

This runs a few example HR questions and prints the assistant responses.

🌐 Streamlit app

streamlit run app.py

Then open the local URL shown in the terminal and ask policy-related questions in the browser.

💡 Usage Example

Example questions you can ask:

• How many paid leave days do I get?
• What is the notice period during probation?
• Can I work from home every day?
• What are the employee reimbursement rules?

📝 Notes

• The first run may take longer because the vector store is being built from the policy file.
• The generated FAISS index is stored in data/faiss_index/ so future runs can load it quickly.
• The assistant is designed to answer from retrieved policy evidence and avoid guessing when the information is not present.

📜 License

This project is licensed under the MIT License. See LICENSE for details.

🤝 Contributing

Contributions are welcome. If you want to improve the assistant, add new features, or make the retrieval pipeline more robust, feel free to open a pull request with a clear description of the change.

🙏 Acknowledgements

• LangChain for orchestration and agent tooling
• Groq for LLM inference
• Jina for embedding models
• FAISS for vector similarity search
• Streamlit for the interactive interface