How k8pil/ttt25-private works
Project Type & Purpose
• Career Conquer is a monolithic full-stack Flask web application for career guidance and interactive interview practice.
• It combines REST APIs and server-rendered pages (MVC pattern), machine-learning body-language analysis, OCR resume parsing, and recommendation engines.
Main Architectural Components
a. Client Browser (Frontend)
• UI: HTML templates (templates/*.html)
• Static assets: CSS/JS (static/, app/static/)
• Webcam & JS modules (guidance.js, interview-chat.js) for real-time video capture.
b. Web Server (Flask Application)
• Entry point: app.py / run.py
• Blueprints / Routes:
• main_routes (login/signup/dashboard)
• api_routes (JSON endpoints for AJAX/video data)
• interview_routes (mock interview flows)
• Service layer (app/services & interview_advisor package):
• resume_service: invokes PyMuPDF, Tesseract OCR, file storage
• interview_service: orchestrates video analysis, emotion scoring via MediaPipe
• tts_service: text-to-speech responses
• recommendation engine (improved_career_recommendations.py, interview_advisor/recommendation.py)
• Data models & persistence (app/models/database.py & SQLAlchemy)
c. Data Stores
• SQLite (interview_metrics.db, data/*.sqlite) with SQLAlchemy ORM
• JSON metric files (data/interview_metrics_*.json) for offline analysis
d. External Libraries & Tools
• MediaPipe (body-language detection)
• scikit-learn (career matching classifier)
• PyMuPDF & Tesseract OCR (resume parsing)
• Flask extensions (flask-login, etc.)
e. Supporting Scripts
• manage_eye_db.py / create_eye_database.py: DB migration/scripts
• check_db_tables.py: health checks
• video_analysis.py / audio_to_text.py: standalone processing modules
• run_interview_advisor.py: wrapper to start advisor engine
Component Interactions & Data Flow
• User browser → HTTP(S) → Flask (routes)
• Routes call Services → External libraries (OCR, MediaPipe, ML models) → process data
• Services persist results → Database (SQLAlchemy) + JSON files
• Flask returns rendered templates or JSON payloads → Browser UI updates
• Real-time video frames streamed via WebRTC/JS → Flask API → MediaPipe → feedback via WebSocket/AJAX
Architectural Patterns / Design Principles
• MVC / Blueprint separation: clear split between routes (controllers), services (business logic), models (data layer), templates/static (view layer).
• Service Layer: isolates ML/OCR integrations from controllers.
• Modular Python packages (interview_advisor) for reusable components.
• Single-process monolith, no microservices—simplifies deployment.