Service Details

MalhaeBang (말해방)

Project Duration: Apr 2024 – Present (Ongoing)

AI Chatbot-based Rental Property Recommendation Platform

  • Currently Used Tech Stack:
    Infrastructure: AWS (ACM, ALB, ECR, EKS, S3), Helm, Docker
    Monitoring: Prometheus, Grafana
    Backend Services: Spring Boot, FastAPI, Flask, Django
    Data: AWS RDS (MySQL), Elasticsearch
    Frontend: Thymeleaf, HTML, CSS, JavaScript
    External APIs: Kakao Maps, Social Login (Kakao, Naver, Google)
  • Team: 5 members (My Role – Fullstack / Infra / DevOps, Others – Data (2), AI (2))
  • Logic Design: Malhaebang Logic
  • GitHub: MalhaeBang Organization

📌 Overview
• MalhaeBang is a real-time rental property recommendation platform powered by an AI chatbot.
• It delivers personalized listings through natural language queries from users.

🎯 Goals & Features
• Real-world rental recommendation service deployed on AWS EKS
• AI chatbot provides tailored listings based on natural language input
• Image-based similar property recommendations
• Microservice architecture for scalability and independence

🚀 Live Services
• Web Server: Spring Boot main application
• AI Chatbot: FastAPI-based NLP interaction service
• Similar Listings: Flask-based recommendation engine using FAISS vector search
• Saju Service: Django-based traditional fortune-based property recommender
• News Service: Real estate news crawler deployed via K8S CronJob

💻 My Role & Contributions
• Fullstack development: Spring Boot APIs, FastAPI chatbot, Thymeleaf frontend
• AWS infrastructure: EKS cluster setup & management
• Monitoring: Prometheus + Grafana-based system health monitoring
• Containerization: Docker optimization and Kubernetes deployment
• Security: Spring Security, HTTPS (ACM), IAM access control

🛠️ Major Troubleshooting
1. EKS Pod Resource Optimization
• Issue: Frequent restarts in RecommendSimilar Pod due to memory exhaustion
• Cause: FAISS index requires >3GB memory
• Solution: Introduced PVC cache, increased memory to 3Gi
• Result: Stable AI recommendation without service restarts

2. S3 Migration for Large-Scale Images
• Issue: 315,359 images relied on Naver CDN; high failure rate under load
• Solution: Migrated to AWS S3 bucket
• Result: Search latency reduced under 1 sec; UX greatly improved

3. OAuth2 Authentication Stability
• Issue: Errors during re-registration of withdrawn users via social login
• Solution: Added user status validation and exception handling
• Result: Reduced auth errors by 95%; improved user flow


✅ Achievements
• High availability infrastructure with AWS EKS
• Orchestration of 8 independently scalable microservices
• Real-time health checks using Grafana dashboards
• Kubernetes-native auto-scaling for traffic fluctuations
• Strong security: ALB, HTTPS, OAuth social login integration