JavaFX 17 · Maven · MySQL (Aiven Cloud) · Advanced AI (Gemini, Groq, Ollama) · Biometrics · Python STT & DeepFace
An enterprise-grade, state-of-the-art digital banking desktop application designed to redefine client interactions, branch management, and back-office banking operations. Powered by a robust JavaFX 17 architecture, Aiven Cloud Databases, and advanced Artificial Intelligence, the platform provides a unified workspace segmented into three dedicated spaces: Client Space, Bank Agent Space, and Platform Administrator Space.
Important
Showcase Portfolio Notice: For intellectual property reasons, the source code of this repository is private. This documentation and screenshots serve as an architectural showcase. For access requests, collaborations, or a live demo, contact the author.
- Developer: Mehdi Mejri
- Email: mehdimejri15@gmail.com
- LinkedIn: linkedin.com/in/mehdi-mejri
The desktop client separates interface rendering from remote services, communicating with cloud components and external API gateways:
graph TD
UI([JavaFX GUI Views]) --> Router{JavaFX Controller Manager}
Router -->|Client View| ClientSpace[Client Space - ROLE_CLIENT]
Router -->|Agent View| AgentSpace[Agent Space - ROLE_AGENT]
Router -->|Admin View| AdminSpace[Admin Space - ROLE_ADMIN]
subgraph Python_Services [Python Native AI Bridges]
DF[DeepFace Biometrics - face_login.py]
Whisper[Google STT / Whisper Engine - stt.py]
end
subgraph Advanced_Java_Services [Advanced Business Logic & Services]
Groq[Groq AI Service - llama-3.1-8b-instant]
Gemini[Gemini 2.5 Flash Service]
Ollama[Ollama Service - Local Llama3 / Mistral]
Ocr[OcrSpace Service]
PdfShift[PdfShift Conversion API]
VoiceRSS[VoiceRSS Text-To-Speech API]
Leaflet[Leaflet.js & Nominatim Geocoder]
Twilio[Twilio SMS Gateway]
POI[Apache POI Excel Service]
Stripe[Stripe Payments API]
end
ClientSpace --> Advanced_Java_Services
AgentSpace --> Advanced_Java_Services
AdminSpace --> Advanced_Java_Services
ClientSpace --> Python_Services
Advanced_Java_Services --> DB[(Aiven Cloud MySQL)]
The relational schema of the entities is designed to support the 5 core banking business domains:
classDiagram
direction TB
class Utilisateur {
+int id
+string nom
+string prenom
+string email
+string telephone
+string password
+string roles
+string statutCompte
+int banqueId
}
class Profile {
+int id
+int userId
+string niveauExperience
+string preferences
+string photo
+string bio
}
class Banque {
+int id
+string nomBq
+string siteWeb
+string telephoneBq
+string emailBq
+string logo
+string statut
+string description
}
class Agence {
+int id
+int banqueId
+string nomAg
+string adresse
+string ville
+string codePostal
+string telephone
+string email
+string horaireOuverture
+double latitude
+double longitude
}
class Service {
+int id
+string nomService
+string description
+int dureeEstimee
+bool disponible
+decimal frais
+string documentsRequis
+string categorie
+string prioriteDefaut
}
class RendezVous {
+int id
+int clientId
+int agentId
+int serviceId
+int agenceId
+int banqueId
+LocalDate dateRdv
+LocalTime heureRdv
+string statut
+string ticketReference
}
class Financement {
+int id
+int clientId
+int agentId
+int offreId
+string typeDmd
+string statut
+decimal montant
+int scoreIa
+string decisionIa
+LocalDateTime dateDecisionIa
}
class Document {
+int id
+int financementId
+string nomFichier
+string typeDocument
+string cheminStockage
}
class TraitementDocument {
+int id
+int financementId
+int score
+string decision
+string justification
+string detailsCalcul
+LocalDateTime dateDecision
}
class Offre {
+int id
+int banqueId
+string nom
+string description
+Date dateDebut
+Date dateFin
+float tauxInteret
+decimal montantMax
+decimal montantMin
}
class Conditions {
+int id
+int offreId
+float tauxSpecial
+decimal montantSeuil
+int dureeMax
}
class Badge {
+int id
+string code
+string label
+string icon
+string description
}
class UserBadge {
+int id
+int userId
+int badgeId
+LocalDateTime earnedAt
}
Utilisateur "1" *-- "1" Profile : possède
Utilisateur "1" *-- "*" UserBadge : débloque
UserBadge "*" o-- "1" Badge : référence
Banque "1" *-- "*" Agence : possède
Banque "1" *-- "*" Offre : propose
Banque "1" *-- "*" Service : offre
Offre "1" *-- "*" Conditions : régit
RendezVous "*" --> "1" Utilisateur : avec (client/agent)
RendezVous "*" --> "1" Service : requiert
RendezVous "*" --> "1" Agence : chez
Financement "*" --> "1" Utilisateur : sollicité par (client)
Financement "*" --> "1" Offre : concerne
Financement "1" *-- "*" Document : contient
Financement "1" *-- "1" TraitementDocument : évalué par
The system's capabilities are divided into five functional modules, each featuring advanced integrations and business rules:
- Facial Biometrics Login: Combines OpenCV image capture in Java (
FaceRecognitionService) with DeepFace Python backend (face_login.py). Computes real-time face comparisons using a pre-calibrated grayscale absolute difference (Core.absdiff) as well as DeepFace Facenet verification. - OAuth2 Authentication Server Hook: Initiates authentication with Google or GitHub APIs by spawning a lightweight local HTTP Server (
com.sun.net.httpserver.HttpServer) in Java on ports9090and9191to dynamically capture access token callbacks. - Smart reCAPTCHA Integration: Implements Google reCAPTCHA v2 inside a JavaFX WebEngine. Capable of communicating checkbox status back to the parent Java runtime via JS bridges.
- Voice-Assisted Registration (Vosk-FR / NLP): Direct onboarding through dictation. Processes speech signals via Vosk offline models or external engines to automatically fill registration fields, with phonetic normalization for domains.
- Gamified Badges Engine: Allocates dynamic visual milestones (Standard ⭐ and VIP 👑) based on savings habits, appointment scheduling, and loan actions.
- Dual-Engine Document Verification: Analyzes files (CNIs, salary slips, utility bills) uploaded via client wizards. Runs a normalization parser to strip currency signs, correct OCR errors (e.g.
oooto000), and check formats. - Multi-Tier Automated Scoring (Out of 100): Calculates candidate reliability:
- 30 Points: Document set completeness check (Auto vs Pro requirements).
- 30 Points: Document structural OCR validity (detecting keywords like
NET À PAYER). - 20 Points: Presence of essential fields (CNI digits, net income values, utility addresses).
- 20 Points: Coherence delta checking (salary age < 3 months, Tunisian identity pattern matches).
- Explainable AI (XAI) Engine: Leverages Groq's
llama-3.1-8b-instantto automatically transform raw numeric scores and failure logs into plain-text decision sheets (Valide, À Corriger, Rejeté) complete with structured tips. - Stripe Payment Gateway: Integrates Stripe's SDK to enable payment of loan processing fees (
Frais de dossier) during application submissions.
- Dynamic Amortization Models: Simulates Mixed, Variable (e.g., changes after month 24), and Fixed schedules, calculating monthly breakdowns and TAEG.
- Live Market Rate Sync: Fetches exchange rates (EUR→USD via ExchangeRate-API, USD→EUR via exchangerate.host) with cached http clients to automatically offset internal loan rate recommendations.
- Interactive Ollama Chatbot: Local Mistral/Llama3 chatbot running via Ollama REST bindings. Validates duration requirements, amount bounds, and guides the user to the closest matching promotional loan catalog product.
- PDFShift Cloud Exporter: Transcribes simulated terms into A4 documents via PDFShift API. Embeds CSS data graphics, schedules, and custom watermarks.
- OSM leafleting & Nominatim Geocoding: Displays interactive maps (
agency-map.html) inside JavaFX WebView using Leaflet.js. Converts plain-text addresses to Tunisian geographic coordinates using the OpenStreetMap Nominatim Client. - Branch Catalog Controls: AJAX-style status toggles that updates database state on the fly.
- Speech Synthesizer (TTS): Uses Windows PowerShell SAPI Speech Synthesis to read bank details and descriptions dynamically in French without UI threads freezing.
- Overlap-Free Guichet Allocation: A scheduler algorithm that sweeps database calendar time intervals (
StartA < EndB AND EndA > StartB) to allocate branch counter resources without conflicts. - visio Jitsi Meet Integration: Instantly spins up Jitsi visual rooms
https://meet.jit.si/BeelineRDV-{rdvId}-{clientId}for remote consultations. - iCalendar (.ics) Sync Export: Generates RFC-5545 calendar files. Spawns OS-level calendar processes (Outlook, Google Calendar, OS X Calendar) using native Java Desktop opening.
- Gemini 2.5 Flash CRM Copilot: Natural Language interface that acts as an Agent Assistant. Understands prompts like "Annuler le rendez-vous #4" or "Rapport du service automobile", executing SQL commands and replying in styled Markdown.
- Dashboard Hub: View accounts, active loans, badges, and today's branch visits.
- Loan Wizard: Upload documents, execute OCR, check score, pay files via Stripe, view XAI feedbacks.
- Credit Chatbot: Converse with local Ollama to find matching financing options.
- Appointment Booker: Check time-slots, preview branch Leaflet map, print QR tickets.
- CRM Console: Monitor branch metrics, client applications, and appointment schedules.
- FullCalendar Agenda: WebView grid displaying appointments. Supports interactive modal controls.
- Gemini CRM Copilot: Chat interface to schedule, cancel, print PDF reports, or export calendar schedules.
- Catalog Manager: CRUD editing for services, interest rates, and promotions.
- Platform Governance: Distribution statistics, active user counters, and transactional charts.
- Registration Review: Onboard banking agents, check branches credentials, approve or block users.
- Central Database Manager: Review catalogs, offers list, and loan logs.
🎯 Live screenshots of the three workspaces in action
.png) 🔑 Smart Login & Biometrics |
.png) 🎙️ Voice-Assisted Registration |
.png) 📧 SMTP Code Recovery |
.png) 🏠 Client Dashboard Hub |
.png) 👑 Gamified Badges System |
.png) 📊 Dynamic Credit Simulator & Taux |
.png) 🏦 Other Banks & TTS Speaker |
.png) 📋 Multi-Step Loan Application |
.png) 🔍 Document OCR Analysis |
.png) 🧠 Explainable AI Score & Decison |
.png) 📅 Intelligent Booking & Map Leaflet |
.png) 📹 Visio-Conference & PDF QR Ticket |
.png) 🏢 Agent CRM Dashboard |
.png) 📆 FullCalendar WebView Sync |
.png) 🤖 Gemini CRM Copilot |
.png) 🧰 Services Catalog Editor |
.png) 🛡️ Admin Overview & Analytics |
.png) 👥 User Approvals & Access Controls |
.png) 📋 Agent Onboarding Queue |
- Core SDK: Java 17 (OpenJFX 17.0.6)
- Build System: Maven 3.8+
- Database Driver:
mysql-connector-j(8.0.33) - Secure Hasher:
jbcrypt(0.4) - API clients: Java HTTP Client (OAuth2, Nominatim, PDFShift, VoiceRSS)
- Data Serializer: Google Gson, Jackson Databind, org.json
- SMS Gateway: Twilio SDK (9.14.0)
- Mail Server: jakarta.mail (2.0.1)
- QR Generator: Google ZXing (3.5.3)
- Document Editors: iText PDF (5.5.13.3) & Apache POI (5.2.3)
- Python Integrations:
- OpenCV Python & DeepFace (Facenet Model)
- SpeechRecognition Python (Google/Whisper models)
- Java Development Kit (JDK): Install JDK 17.
- Maven: Ensure Maven is configured in your systems environment.
- Python Environment: Create a local environment and install required modules:
python -m venv face_env source face_env/Scripts/activate pip install deepface opencv-python speechrecognition - Offline Vosk Model: Download the French model from Vosk and save it under
models/vosk-fr.
Build and run the JavaFX runtime:
mvn clean compile
mvn javafx:runThe JDBC url connects to the shared Cloud MySQL cluster using SSL Mode:
jdbc.url=jdbc:mysql://mysql-pi-project-pi1.j.aivencloud.com:26903/beeline_db?useSSL=true&requireSSL=true&sslMode=REQUIRED
jdbc.username=avnadmin
jdbc.password=YOUR_DB_PASSWORDTo load database assets, import the SQL script located in:
Dump/version-Full_Integration4-beeline_db-202603010016.sql
Made with ❤️ by Mejri Mehdi