🧠 CV Insight AI

CV Insight AI is an end-to-end Streamlit-based application designed to intelligently analyze, score, summarize, and conduct Q&A on a batch of candidate CVs.

This app runs entirely locally, integrating Semantic Text Matching, a GGUF-based LLM (via Llama.cpp), and Retrieval-Augmented Generation (RAG) using a Faiss vector database.

💡 Parsing & Scoring Algorithm Details: For an in-depth explanation of how the CV parsing and semantic scoring processes work, see the Kaggle notebook: End-to-End CV Parsing and Semantic Scoring

---

✨ Key Features

1. Automated CV Parsing: Reads and extracts essential information from PDF files (Experience, Education, Skills, etc.).
2. Semantic Scoring: Calculates a match score between the candidate's CV and the input Job Title, Job Description, and Required Skills.
3. AI Summary (Pros & Cons): Generates a concise summary highlighting the Strengths and Weaknesses of each shortlisted candidate using a local LLM.
4. Chat with Candidates (RAG Q&A): Ask free-form questions about shortlisted candidates. The AI retrieves context from the vector database (Faiss) and answers based purely on the CV data.

---

🛠️ Tech Stack & Requirements

This application is built with Python 3.10 and a fully pinned dependency stack to ensure stability across ML, RAG, and local LLM workflows.

🧠 Core Machine Learning & NLP

• torch==2.0.1
• transformers==4.30.2
• sentence-transformers==2.2.2 (used with the all-MiniLM-L6-v2 embedding model)
• huggingface_hub==0.14.1
• numpy==1.26.4 (pinned to avoid binary incompatibility with PyTorch and FAISS on Windows)

🔍 Retrieval-Augmented Generation (RAG)

• faiss-cpu==1.13.2 (vector similarity search for document retrieval)
• llama-cpp-python==0.3.16 (efficient local inference for GGUF LLM models)

📄 Data Processing

• pandas==2.3.3
• scikit-learn==1.7.2
• PyMuPDF==1.26.7 (PDF text extraction and parsing)

📊 UI & Visualization

• streamlit==1.54.0 (web-based UI / frontend)
• plotly==6.5.2

🛠 Utilities

• python-dotenv==1.2.1 (environment variable management)
• requests (HTTP requests for external integrations)

⚠️ Environment Notes

• Python 3.10 is required. Other versions may cause runtime or binary issues.

• All dependencies are version-pinned to prevent:

  • NumPy / PyTorch ABI mismatches
  • FAISS runtime errors
  • Incompatibilities with local LLM inference

• Tested on Windows (CPU-only) environments.

• Use requirements.txt for installing dependencies in a fresh virtual environment.

---

🚀 Installation & Setup

1. Clone the Repository

git clone https://github.com/your-username/cv-insight-ai.git
cd cv-insight-ai

2. Download the LLM Model (GGUF)

1. Download a GGUF model from HuggingFace (e.g., gemma-3-4b-it-q4_0.gguf).
2. Create a folder named llm in the project root.
3. Save the downloaded .gguf file into the llm/ folder.

3. Configure Environment Variables (.env)

Create a .env file in the project root and specify the path to your GGUF model:

MODEL_PATH=llm/gemma-3-4b-it-q4_0.gguf

---

💻 Running the App

1. Double-click run.bat (Windows) to auto-setup the virtual environment and install dependencies.
2. On first run, the script will take longer as it creates venv and installs all libraries from requirements.txt.
3. The Streamlit app will open at http://localhost:8501.
4. Subsequent runs will launch instantly since the environment is already prepared.

---

📖 Usage Guide

1. Input Job Data: Enter Job Title, Job Description, Required Skills, and Highlight Keywords in the sidebar. Adjust weights as needed.

2. Choose CV Input Mode:

   • Upload PDFs: Select individual CV files.
   • Select Folder: Batch-process multiple CVs from a local folder.

3. Start Analysis: Click "Analyze CVs". The system parses, scores, visualizes (Radar & Bar Charts), and generates AI summaries.

4. Q&A (RAG): At the bottom of results, type questions like "Who is the most suitable candidate to handle SQL databases?" for AI recommendations based on CV data.

---

Built with ❤️ using Python & Streamlit.