💻 Code | 🤗 Model | 🤖 Demo | 📑 Technical Report
English | 中文
## Project Introduction SmartResume is an layout‑aware resume parsing system. It ingests resumes in PDF, image and common Office formats, extracts clean text (OCR + PDF metadata), reconstructs reading order with layout detection, and leverages LLMs to convert content into structured fields such as basic info, education, and work experience. [demo](https://github.com/user-attachments/assets/db06bf23-9700-4a47-87e6-15ca526ad269) ## Quick Start ### Requirements - Python >= 3.9 - CUDA >= 11.0 (optional, for GPU acceleration) - Memory >= 8GB - Storage >= 10GB ### Installation 1. **Clone the repository** ```bash git clone https://github.com/alibaba/SmartResume.git cd SmartResume ``` 2. **Create conda environment** ```bash conda create -n resume_parsing python=3.9 conda activate resume_parsing ``` 3. **Install dependencies** ```bash pip install -e . ``` 4. **Configure environment** ```bash # Copy configuration template cp configs/config.yaml.example configs/config.yaml # Edit configuration file and add API keys vim configs/config.yaml ``` ### Basic Usage #### Method 1: Command Line Interface (Recommended) ```bash # Parse single resume file python scripts/start.py --file resume.pdf # Specify extraction types python scripts/start.py --file resume.pdf --extract_types basic_info work_experience education ``` #### Method 2: Python API ```python from smartresume import ResumeAnalyzer # Initialize analyzer analyzer = ResumeAnalyzer(init_ocr=True, init_llm=True) # Parse resume result = analyzer.pipeline( cv_path="resume.pdf", resume_id="resume_001", extract_types=["basic_info", "work_experience", "education"] ) print(result) ``` > **Tip**: The CLI reloads models on every call, suitable for one-off use. For repeated parsing, use one of the following to avoid reloading: #### Web Service (Recommended for repeated use) ```bash python demo/web_app.py # Open http://localhost:4999 in your browser to upload resumes ``` #### Batch Processing ```python from smartresume import ResumeAnalyzer import glob, json analyzer = ResumeAnalyzer(init_ocr=True, init_llm=True) # load once for f in glob.glob("/path/to/resumes/*.pdf"): result = analyzer.pipeline(cv_path=f, resume_id=f.split("/")[-1]) print(json.dumps(result, ensure_ascii=False, indent=2)) ``` ### Local Model Deployment SmartResume now supports local model deployment using vLLM, reducing dependency on external APIs: ```bash # Download Qwen-0.6B-resume model python scripts/download_models.py ``` Configure `use_direct_models: true` and `direct_model_name` in `configs/config.yaml`; no separate vLLM server is required (vLLM runs in-process). For detailed local model deployment guide, see [LOCAL_MODELS](docs/local-models.md). ## Key Features | Metric Category | Specific Metric | Value | Description | |----------------|----------------|-------|-------------| | **Layout Detection** | mAP@0.5 | **92.1%** | High layout detection accuracy | | **Information Extraction** | Overall Accuracy | **93.1%** | High accuracy | | **Processing Speed** | Single Page Time | **1.22s** | High performance | | **Language Support** | Supported Languages | **many** | Covering major global languages | ### Benchmark Results For detailed benchmark results, see [Benchmark Results](docs/benchmark-results-en.md). ## Configuration For detailed configuration options, see the [Configuration Guide](docs/configuration.md). ## License Information This project is licensed under the [LICENSE](LICENSE). We plan to explore and replace them with models under more permissive licenses to enhance user-friendliness and flexibility. ## Important Notice Due to open-source compliance requirements, this codebase is a refactored version. The internal PDF parsing and OCR components cannot be published. We have replaced them with open-source alternatives, and some features may not be fully compatible with the original implementation. ## Acknowledgments - [PDFplumber](https://github.com/jsvine/pdfplumber) - [RapidOCR](https://github.com/RapidAI/RapidOCR) ## Citation ```bibtex @article{Zhu2025SmartResume, title={Layout-Aware Parsing Meets Efficient LLMs: A Unified, Scalable Framework for Resume Information Extraction and Evaluation}, author={Fanwei Zhu and Jinke Yu and Zulong Chen and Ying Zhou and Junhao Ji and Zhibo Yang and Yuxue Zhang and Haoyuan Hu and Zhenghao Liu}, journal={arXiv preprint arXiv:2510.09722}, year={2025}, url={https://arxiv.org/abs/2510.09722} } ``` --- **Note**: Please ensure compliance with relevant laws and regulations and privacy policies.