Skip to content

NEXXT AI Hackathon - Raiffeisen Bank. Top 7 finalist project with a core idea of - Banks lose customers not because of poor service, but due to lack of personalized engagement. Our system keeps clients connected to their bank through intelligent, individualized content and proactive risk management.

Notifications You must be signed in to change notification settings

floreaGabriel/NEXXT_AI_PROJECT

Repository files navigation

🏦 NEXXT AI Hackathon 2025 - Customer Retention Intelligence System

Team AI Agents Model Cloud ML Hackathon

📋 About The Project

Developed by Team ATM during the NEXXT AI Hackathon 2025 (November 1-2), this intelligent system revolutionizes customer retention in banking through personalized AI-driven experiences. Selected from 250+ applicants, our solution addresses one of banking's most critical challenges: keeping customers engaged with their financial institution.

🎯 Core Problem

Banks lose customers not because of poor service, but due to lack of personalized engagement. Our system keeps clients connected to their bank through intelligent, individualized content and proactive risk management.

💡 Our Solution

An AI-powered Customer Retention Intelligence System that combines:

🔹 Personalized Content Engine - Dynamic, user-specific banking content based on comprehensive user profiles
🔹 Churn Risk Prediction - Bayesian probabilistic models calculating each customer's likelihood of leaving
🔹 Multi-Agent AI Orchestration - 13 specialized LLM agents working collaboratively via OpenAI SDK
🔹 Enterprise AI Infrastructure - Claude 4.5 Sonnet via AWS Bedrock for production-grade reliability

✨ Key Features

🤖 13-Agent AI Architecture

Our system employs 13 specialized LLM agents using the OpenAI Agents SDK, each focused on specific customer engagement tasks:

  • Profile Analyzer Agent - Deep user behavior analysis
  • Content Personalization Agent - Tailored banking insights
  • Risk Assessment Agent - Churn prediction coordination
  • Communication Agent - Optimal messaging strategies
  • Product Recommendation Agent - Context-aware financial products
  • Sentiment Analysis Agent - Customer satisfaction monitoring
  • Engagement Optimizer Agent - Interaction timing and frequency
  • Financial Health Agent - Personalized financial wellness tips
  • Transaction Pattern Agent - Spending behavior insights
  • Lifecycle Management Agent - Customer journey optimization
  • Retention Strategy Agent - Proactive intervention planning
  • Feedback Processing Agent - Continuous improvement loop
  • Orchestrator Agent - Multi-agent coordination and workflow

📊 Bayesian Risk Scoring

Advanced probabilistic modeling calculates churn risk scores for each customer:

P(churn | user_features) = Bayesian_Model(
transaction_freque
cy, engagement_
etrics, pr
duct_usage, customer
lifetime_value, i
teraction_patter

🎨 Personalization Engine

Every customer receives content tailored to their:

  • Financial goals and aspirations
  • Transaction history and spending patterns
  • Product usage and preferences
  • Risk profile and life stage
  • Communication preferences
  • Engagement behavior

🛠️ Technology Stack

  • AI Framework: OpenAI Agents SDK
  • Language Model: Claude 4.5 Sonnet (via AWS Bedrock)
  • ML Libraries: Python, scikit-learn, PyMC (Bayesian modeling)
  • Data Processing: Pandas, NumPy, Jupyter Notebooks
  • Cloud Infrastructure: AWS Bedrock
  • Development: Jupyter Notebook (92.5%), Python (7.5%)

👥 Team ATM

  • Florea Cristian Gabriel (@floreaGabriel) - AI Architecture & Integration
  • Stan Sabin (@SabinGhost19) - AI Architecture & Integration
  • Luca-Teodor Apahidean (@lucaapahidean) - AI Architecture & Integration
  • Gaitan Nicolai (@Nicky261) - Machine Learning & Risk Models

🏆 Hackathon Context

  • Event: NEXXT AI Hackathon 2025
  • Sponsor: Raiffeisen Bank Romania
  • Duration: 36 hours (November 1-2, 2025)
  • Location: Commons Unirii, Bucharest
  • Competition: 13 teams from 250+ applicants
  • Focus: AI Innovation in Banking & Financial Services

📊 System Architecture

┌─────────────────────────────────────────────────┐
│ User Profile Database                           │
│ (Demographics, Transactions, Behaviors)         │
└────────────────┬────────────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────────────┐
│ Bayesian Risk Scoring Engine                    │
│ (Churn Probability Calculation)                 │
└────────────────┬────────────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────────────┐
│ 13 LLM Agents (OpenAI SDK)                      │
│ Claude 4.5 via AWS Bedrock                      │  
│ ┌─────────────────────────────────────┐         │
│ │ Orchestrator Agent (Coordinator)    │         │
│ └──────────────┬──────────────────────┘         │
│                │                                │
│ ┌──────────────┼──────────────────────┐         │
│ │ Profile │ Content │ Risk │ Comms    │         │
│ │ Product │ Sentiment│ Engagement     │         │
│ │ Transaction │ Lifecycle │ Retention │         │
│ │ Feedback │ Financial Health         │         │
│ └─────────────────────────────────────┘         │
└────────────────┬────────────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────────────┐
│ Personalized Content & Interventions            │
│ (Delivered to Customer)                         │
└─────────────────────────────────────────────────┘

📈 Results & Impact

  • Churn Prediction Accuracy: Bayesian model with probabilistic confidence intervals
  • Personalization Coverage: 100% of user profiles
  • Agent Response Time: Real-time inference via AWS Bedrock
  • Customer Engagement: Dynamic content adaptation based on user behavior

💼 Business Value

  1. Reduced Churn Rate - Proactive identification of at-risk customers
  2. Increased Engagement - Personalized content drives daily interactions
  3. Higher Customer Lifetime Value - Stronger bank-customer relationships
  4. Operational Efficiency - Automated personalization at scale
  5. Data-Driven Insights - Continuous learning from customer behavior

🔮 Future Enhancements

  • Real-time streaming analytics for instant risk updates
  • Multi-language support for international banking
  • Integration with mobile banking apps
  • A/B testing framework for content strategies
  • Advanced sentiment analysis from customer communications
  • Predictive product recommendation engine

🙏 Acknowledgments

  • Raiffeisen Bank Romania for sponsoring and hosting the hackathon
  • NEXXT Summit organizers for the incredible event infrastructure
  • AWS for cloud computing resources
  • Anthropic for Claude 4.5 via Bedrock
  • All mentors and advisors who guided our development
  • Fellow competing teams for inspiring collaboration

📝 License

This project was developed during the NEXXT AI Hackathon 2025.

📧 Contact

For questions, collaboration, or implementation inquiries:

  • Open an issue in this repository
  • Connect with team members on LinkedIn

If you find this project interesting, please star this repository!


Built with ❤️ during 36 hours of intense innovation

#NEXXTAIHackathon #AIInnovation #MachineLearning #FinTech #CustomerRetention

About

NEXXT AI Hackathon - Raiffeisen Bank. Top 7 finalist project with a core idea of - Banks lose customers not because of poor service, but due to lack of personalized engagement. Our system keeps clients connected to their bank through intelligent, individualized content and proactive risk management.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •