Resort Booking System: Backend & Infrastructure Refactor
Backend & Infrastructure
.webp)
An architectural overhaul and AI modernization of a resort booking system originally developed by a 5-member team. Built on top of the core Django REST backend, this evolution introduced Redis distributed locking for race-condition prevention, sub-millisecond query optimization, Celery-driven PayMongo automation, and a multi-turn LangGraph booking agent with a RAG FAQ concierge hosted on AWS.
- 01Eliminated double-booking race conditions using Redis distributed locks during room selection, while offloading PayMongo webhook processing and automated confirmation emails to an asynchronous Celery queue.
- 02Reduced database queries by 78% to 99.5% across 10 endpoints by eliminating N+1 query patterns with batch prefetching, dropping worst-case queries from 2,103 down to 12 on a 100-record benchmark.
- 03Migrated from DigitalOcean PaaS to AWS (IaC) by architecting a custom VPC with separated public/private subnets, containerizing services via Docker Compose on EC2, and migrating data persistence to AWS RDS PostgreSQL.
- 04Built a multi-turn conversational booking agent and RAG hotel concierge on top of the backend using LangGraph, Google Gemini/OpenAI, and Pydantic structured output extraction for deterministic state management.
- 05Implemented full observability, multi-turn state testing, and latency/token tracing for the AI agent pipelines using LangSmith.
AWS & System Architecture
Containerized Django services, Celery workers and Redis database hosted on AWS EC2 within a custom VPC, routed through an Nginx reverse proxy. Sensitive workloads and data persistence are isolated in private subnets via AWS RDS PostgreSQL, while LangGraph and LangSmith orchestrate and monitor the conversational AI booking layer.