How rafacalassara/jobapplicationflow works
Project Type and Purpose
• This is an AI-driven, orchestrated pipeline service with a minimal frontend served by Gradio.
• Purpose: Given a user’s LinkedIn PDF resume, a job posting URL, and optional inputs, the system runs three “crews” of AI agents (resume tailoring, company research, email writing) to generate/output:
• A tailored resume (MD + PDF)
• A company research report
• A job-application email draft
High-Level File-Structure Analysis
• Top-level files & folders:
• app.py, main.py → entrypoints (Gradio UI & orchestration)
• .env.example, requirements.txt → configuration/dependencies
• Dockerfile, docker-compose.yml → containerization/deployment
• inputs/ → raw input templates (e.g., base_resume.md)
• crews/ → 3 sub-modules (tailor_resume_crew, companies_research_crew, email_writer_crew), each with:
– <crew>.py (crew implementation)
– config/agents.yaml, config/tasks.yaml (YAML-driven agent definitions)
• tools/ → helper scripts (PDF conversion, HTML styling, custom tools)
Extract Key Components
a. Frontend/UI
• Gradio interface defined in app.py
• Hosts web server at http://127.0.0.1:7860
b. Orchestrator Layer
• main.py (or embedded in app.py) that wires together UI inputs, crew invocations, and outputs
c. Crew Modules (Pipeline Stages)
• Tailor Resume Crew
• Companies Research Crew
• Email Writer Crew
Each crew reads its config, invokes AI agents via crewAI, and returns structured outputs.
d. Tools Library
• convert_resume_to_pdf.py
• convert_to_styled_html.py
• custom_tool.py
Used by crews to post-process raw AI outputs.
e. External Services
• OpenAI API (via crewAI)
• Serper API (for web/company scraping)
f. Data Stores / File System
• inputs/ for source files
• outputs/ (implicitly created at runtime) for final artifacts
g. Deployment
• Dockerfile (build image)
• docker-compose.yml (service orchestration)
Relationships & Data Flows
• User → Gradio UI (app.py)
• UI passes inputs → Orchestrator (main.py)
• Orchestrator sequentially triggers:
1. Tailor Resume Crew → uses LinkedIn PDF + job URL → outputs tailored MD resume
2. Companies Research Crew → uses job/company info → outputs company report
3. Email Writer Crew → uses resume + report → outputs draft email
• Each Crew may call:
• Tools library for transformations
• External APIs (OpenAI / Serper)
• Final artifacts (MD, PDF, email) are written to outputs/
Architectural Patterns & Design Principles
• Modular, pipeline-oriented design: each “crew” is an isolated stage
• Configuration-driven: YAML agent/task definitions allow swapping or extending behavior without code changes
• Single-process, service-oriented monolith (no microservices)
• Separation of concerns: UI vs. orchestration vs. AI logic vs. tooling
Technologies & Frameworks
• Python 3.12
• crewAI (for agent orchestration)
• Gradio (frontend)
• Docker & Docker-Compose (deployment)
• YAML (configuration)
• Third-party APIs: OpenAI, Serper