Status: Ongoing
Associated with: Thynk360
Recruitment and talent acquisition increasingly rely on automated systems to handle large volumes of resumes. To support this trend, I am developing a Resume Parser & Skill Extractor—an AI tool that automates the process of extracting key candidate information from resumes. This project, part of the Thynk360 initiative, simplifies how recruiters scan, evaluate, and shortlist resumes by structuring unstructured data using NLP.
Tools & Technologies Used
- Programming Language: Python
- Libraries & Frameworks: spaCy, PDFMiner, PyMuPDF (fitz), Regex, Pandas
- NLP Techniques: Named Entity Recognition (NER), Tokenization, Lemmatization
- Output Formats: JSON and CSV for easy integration with ATS systems
Description
This system is designed to process a batch of resumes—mostly in PDF format—and extract structured data such as name, email, phone number, education, work experience, and skills. The parser reads resumes using PDFMiner and PyMuPDF, then cleans and tokenizes the text.
For skill extraction, I used spaCy’s Named Entity Recognition (NER) capabilities, combined with regular expressions and keyword-based matching. The tool matches terms against a predefined skills database that includes both technical and soft skills (e.g., Python, teamwork, project management).
Each parsed resume is converted into a structured JSON format, and the data is collated into a CSV file—ideal for further processing or direct use in Applicant Tracking Systems (ATS).
Key Highlights
- Built a robust PDF parser capable of handling multi-column, graphically rich resumes.
- Implemented NER-based entity extraction to identify candidate details with high accuracy.
- Extracted customized skillsets using a combination of rule-based filters and NLP.
- Exported structured data into ATS-compatible formats (CSV/JSON).
Learned / Achieved
This project deepened my experience in information extraction—a key NLP application. I learned how unstructured data like resumes require different techniques than traditional datasets. For instance, while a resume might list “Education” under a heading, the format varies dramatically from one document to another, requiring dynamic parsing strategies.
By integrating spaCy’s NER models, I discovered how pre-trained models can be extended and fine-tuned with domain-specific labels to improve accuracy. I also explored combining regex patterns with statistical NLP to improve the detection of contact details, date ranges, job titles, and educational credentials.
I gained a better understanding of resume structure variability and the challenge of building a generalized system that adapts to different designs and content formats. This helped me build more modular and error-tolerant code.
From a deployment standpoint, I learned to write clean, testable functions that can be packaged into a standalone tool or integrated into HR software systems. The ability to convert the output into structured formats like JSON and CSV gave this tool practical value in real-world applications.
Future Plans
My next steps involve training a custom NER model using labeled resume data to increase extraction accuracy for industry-specific terms. I’m also planning to build a Streamlit UI where users can upload resumes and instantly see parsed details.
Another future goal is to integrate the parser with LinkedIn profiles or job portals, so recruiters can automatically match parsed resumes with job descriptions—enabling AI-driven candidate-job fit scoring.
This project reflects my broader goal at Thynk360: using AI to streamline real-world workflows, increase efficiency, and eliminate repetitive manual tasks.
