How ztrimus/resumeflow works
Project Identification
• Name: ResumeFlow – An LLM-facilitated Pipeline for Personalized Resume Generation and Refinement.
• Purpose: Automate the resume generation and refinement process, integrating job posting details with a user’s master resume using large language models (LLMs) and LaTeX for final resume PDF generation.
Main Components
A. User Interface Layer
• Web Application Interface (“WebApp”): Represented by the web_app.py, which exposes the tool as an interactive web service (deployed on Streamlit).
• Command Line Interface: Represented by main.py, providing the pipeline’s entry point for command-line usage.
B. Core Processing Module (zlm Package)
• Initialization & Configuration: The zlm/__init__.py, which sets up the module.
• Data Inputs & Demo Data: The demo_data folder (with user_profile.json and user_resume.pdf) that provides sample inputs.
• Prompts: The prompts directory contains resume_prompt.py and sections_prompt.py – these are key for interacting with LLM APIs by framing requests.
• Schema Definitions: The schemas folder (job_details_schema.py and sections_schemas.py) that defines structured representations of job and resume section data.
• Templates: The templates folder includes the LaTeX resume template (resume.cls and resume.tex.jinja) for generating PDF resumes.
• Utility Functions: The utils directory with modules for:
• Data Extraction (data_extraction.py)
• LaTeX Operations (latex_ops.py)
• LLM Model Integration (llm_models.py)
• Performance Metrics (metrics.py)
• General Utilities (utils.py)
• Configuration Variables: For global settings (variables.py).
C. External Dependencies and Integrations
• LLM API Integration: Interaction with external LLM providers like OpenAI and Gemini Pro. This component is primarily invoked via the AutoApplyModel from the zlm package.
• PDF Generation Workflow: Interaction with system-level packages for LaTeX processing (installed externally on Linux/Mac, as indicated in the README).
• External Scripts & Resources: The resources folder (including demo videos, project reports, and shell scripts like publish.sh) that support documentation and deployment aspects.
D. Package and Dependency Management
• Python Package Management: Indicated by packages.txt, pyproject.toml (for Poetry), and resources/requirements.txt. These highlight dependency management and build configuration.
• Development Configurations: The .devcontainer and .vscode directories provide container and IDE-specific configurations for development and debugging.
• Direct these inputs to the Core Processing Module (zlm package), where:
• Data is extracted and formatted using the utilities.
• Prompts are constructed using predefined patterns (from the prompts modules) and sent to the LLM API.
• Response data is validated and restructured using the schemas.
• The resume is rendered by merging the LLM-processed content with LaTeX templates.
• PDF generation occurs through external LaTeX conversion steps (managed by the latex_ops module).
• The generated resume/cover letter is then provided back to the user (downloaded in the specified folder).
Architectural Patterns and Design Principles
• Layered Architecture: Clearly distinguish between presentation (CLI and WebApp), domain processing (LLM integration, resume generation), and infrastructure layers (PDF conversion, external API calls).
• Modular Design: Use Python modules (prompts, schemas, utils) to ensure separation of concerns – each module handles distinct functionalities.
• External Service Integration: Emphasize the role of third-party LLM providers and the LaTeX engine as external services that the system relies on.