Automating Pre-Hire CV Template Generation Using LLM and Live Preview Interfaces
Overview: This thesis focuses on developing an AI-assisted system that automates the generation of company-standard CVs for external candidates. Currently, recruiters must manually re-enter candidate information into the company’s CV template since the internal system requires an employee ID. This process is repetitive, time-consuming, and error-prone. By applying Natural Language Processing (NLP) and Large Language Models (LLMs), the proposed system will automatically extract and structure relevant data from digital CVs, populate the company template, and display a live editable preview—reducing manual work and improving efficiency.
Description
Key Components
1. Automated Information Extraction
a. Use NLP and LLM models to extract structured information (personal data, education, work experience, skills) from digital CVs in formats such as PDF and Word.
b. Define a standardized JSON schema aligned with the company’s CV template to ensure consistent mapping and formatting.
2. Template Mapping and Live Preview
a. Automatically populate the standardized CV template based on the extracted data.
b. Develop a web-based interface with two panes: editable structured data on the left and a real-time preview of the company CV template on the right.
c. Enable export to both PDF and Word formats using libraries such as python-docx or reportlab.
3. Evaluation and Optimization
a. Evaluate extraction accuracy (precision, recall, F1), time saved compared to manual entry, and recruiter usability through the System Usability Scale (SUS).
b. Expected outcomes: ≥90% F1 score on core fields, ≥50% reduction in processing time, SUS ≥75.
Challenges
· Data Variation: Handling different CV layouts and writing styles while maintaining high extraction accuracy.
· Template Consistency: Mapping extracted data to a fixed structure without formatting errors.
· User Interaction: Designing an intuitive interface for quick verification and editing.