Pickply - R2C Company

Situation:

• The service required rapid iteration to accommodate a growing user base, as well as an independent blog channel to establish SEO and attract B2B customers.

Task:

• Accelerate the app release process bypassing store reviews and build an official company blog to drive inbound marketing and business growth.

Action:

• Proposed and directly integrated hot-updater to enable instant Over-The-Air (OTA) app updates, securing an agile release cycle.
• Developed the official Pickply promotional & B2B blog from scratch using inblog to optimize Search Engine Optimization (SEO).

Result:

• Led the full-stack development of Pickply across versions 1.6.17 ~ 2.7.3, scaling MAU from 10K to 70K and supporting over 450,000 cumulative users.
• Actively drove business growth and user acquisition, contributing to the successful closure of the Pre-A fundraising round.

Situation:

• Manual deployments caused downtime, and infrastructure scaled inefficiently, leading to high operational costs ($650/month) and bottlenecks.

Task:

• Redesign the infrastructure to fully automate deployments, eliminate downtime, and significantly reduce recurring operational and infrastructure costs.

Action:

• Redesigned and deployed a secure AWS infrastructure using CDK and introduced a Blue/Green deployment strategy.
• Automated CI/CD pipelines using Docker multi-stage builds, ECR caching, and CodePipeline.
• Downgraded MongoDB specs (M30 → M10) via indexing/monitoring and introduced Serverless ETL pipelines.

Result:

• Eliminated deployment downtime (7s → 0s) and reduced manual deployment time by 80% (5min → 1min).
• Cut monthly infrastructure costs by over 57% ($650 → $280).
• Maintained Serverless ETL pipeline operations at under $3/month.

Situation:

• The main server and cron scheduling server were divided into separate ALB+EC2 instances, incurring an additional $150+ monthly.
• Exposed to a Single Point of Failure (SPOF) risk with complex HTTP dependencies.

Task:

• Consolidate the separated scheduling server into the main server to reduce costs and remove the SPOF risk, all while ensuring zero service downtime.

Action:

• Migrated to Fargate with zero downtime through gradual weight-based traffic routing (Blue/Green).
• Eliminated dependencies and latency by consolidating HTTP API calls into internal service calls.
• Minimized costs by setting up ECS Fargate scheduled scaling for 15 peak times.

Result:

• Reduced dual operational costs by 50% by eliminating redundant infrastructure and DataTransfer overhead.
• Resolved SPOF risk and reduced deployment/maintenance time by unifying management points.

Situation:

• Configured notification data volatilized during server restarts due to Node.js setTimeout and in-memory operations.
• Queue distributed processing was impossible during auto-scaling, risking redundant execution (Race Condition) upon recovery.

Task:

• Build a highly reliable, persistent scheduling architecture that guarantees 100% execution without data loss during restarts or duplicate runs in multi-server environments.

Action:

• Ensured data persistence by transitioning to a BullMQ-based message queue architecture.
• Implemented Distributed Locks using Redis SET NX EX to block redundant execution.
• Built a @bull-board dashboard to visualize the queue state and introduce automated Retries.

Result:

• Achieved Zero Data Loss for scheduled tasks during deployments and restarts.
• Successfully paralleled task processing across instances, resolving overlap and single-point-of-failure issues.

Situation:

• All 3 nodes scaled up (M20 → M40) during auto-scaling, wasting $300~$400 monthly.
• Heavy queries from analytics and full-scan batches interrupted Primary communications, delaying real-time service operations.

Task:

• Prevent unnecessary scale-ups, isolate performance bottlenecks, and resolve slow queries to drastically reduce database infrastructure costs while ensuring stability.

Action:

• Separated long-running analytics traffic to a ReadOnly (Analytics) Replication Node (Read Replicaarchitecture).
• Optimized 40-second malicious queries into Covered queries using a Compound Index.
• Reduced storage waste by swapping PIT (Point-In-Time) backups to daily snapshot backups.

Result:

• Downgraded database specs and cut monthly DB costs by 70% ($400 → $120).
• Reduced 40-second analytics queries to under 0.1 seconds, completely resolving load bottlenecks.