Multi-Cloud
Data Protection
ValidationFind every failure first
I design and lead system testing for a backup and restore platform on AWS, GCP and Azure. Fault injection, scale testing from 1.6B real objects to 70B synthetic ones, multi-tenant isolation and CI pipelines.
A gate people trust
The scheduled GCS pre-checkin gate, before and after two months of fixing flaky waits, races and environment collisions.
Pre-checkin gate
What I build
What I built as primary owner of the S3 and GCS test areas.
Pipelines people trust
I build and maintain the GCS pre-checkin gate and its stage jobs. Over two months I removed flaky waits, race conditions and environment collisions. The scheduled gate went from 77% to 96% green, and failures fell from 23% to 4%. The arena deploy pipeline holds 98% success.
GCS backup test suite, from zero
As the platform added Google Cloud, I led the whole test program: SDK layer, API wrappers, a daily pre-checkin pipeline, protection groups and policies, versioned restore chains, archive-tier restores, backup filters and point-in-time restore.
S3 backup & restore end to end
Coverage for incremental backups, object filters, storage classes and archive tiers, customer-key encryption, checksums, multipart uploads, malware detection and cross-region restore. Led pre-release automation for instant access and point-in-time bucket rollback.
Fault-injection matrices
Designed helpers and pipelines that break things mid-backup or mid-restore: deleted or overwritten sources, revoked access, requester-pays buckets, legal holds, broken encryption keys. Every test asserts on what the customer actually sees.
Event-driven continuous backup
Validated micro-backups, reprotect flows, carry-over events, soft vs hard deletes and multi-topic topologies on both clouds. Moved the suites from internal triggers to the public customer API.
Multi-tenant isolation (RBAC)
Built a coverage matrix for cross-unit denial, moving resources between units, list scoping and attribution on tasks and billing, and found isolation gaps along the way.
From 1.6B real to 70B synthetic objects
Backed up 1.6B real objects (440 TB) with copy throughput holding at about 115K objects/s from 800M to 1.6B. Then pushed backup and restore to 70B objects with synthetic inventories. Found the bottlenecks (concurrency caps, an inventory wait that took ~80% of wall time, pod CPU, metadata-store limits) and verified each fix.
Scale, lifecycle & metering
A synthetic-event harness that confirmed CDP keeps up at 50M events per window, plus long-running soak suites. Automated expiry and garbage-collection tests and checked usage reports against rate cards.
IaC onboarding & CI
Automated cloud-account onboarding with Terraform and CloudFormation: upgrades, state-backed drift detection, recovery paths. Ran Jenkins pipelines and brought two retired test environments back into service.
Test framework & quality
Built a test-flow reporting engine for CI, a monitor for long-running tasks and REST/gRPC client wrappers. Led audits that found and fixed assertions that could never fail.
Owning the program, growing the team
Primary owner of the S3 and GCS test areas. Led the epics and test plans with exit criteria and requirement traceability. Built the onboarding tutorial and guides for new QE hires, wrote interview exercises and scorecards, and sit on the interview panel.
I run my work through agents
I don't use AI as autocomplete. Every ticket runs through an agentic pipeline I designed, measured and keep tuning.
Measured, not guessed
Codified workflows
Ticket kickoff, PR review guides, Jenkins triage and pre-PR gates are written down as reusable skills. Five read-only subagents do the heavy reading and report back.
An agent that learns
A curated memory of 300+ notes records system quirks, failure signatures and past decisions. Hooks enforce house rules before a commit or PR is written, and cut median answer length by 24%.
Second opinions on every branch
A second model reviews branches in parallel. On one sweep it caught two real defects that the first pass had missed.
Experience
Software Engineer (QE / SDET) · Clumio, a Commvault company
- Lead for GCS data-protection system testing; co-owner of S3 system testing
- Write test plans and fault matrices, break epics into work and drive them
- QE lead for S3 Instant Access (early access to GA) and S3 Rewind snapshot rollback
- Owned QE for the new GCS backup product from zero to GA and beyond
- Wrote the Azure Blob P0 test strategy and built its shared test base (2026)
- Catch product bugs before release: silent data loss, authorization bypasses, scale limits, races. About a quarter are rated High or above
Software Engineer · IDCITI
- uGPS: seamless GNSS across a 1.5 km subway tunnel (MobiCom 2022)
- Indoor drone-testing solution; IDefense
B.S. / M.S. Computer Science (combined program) · Stony Brook University
- Mobile Computing & Wireless Networks
Papers & patents
- uGPS: Design and Field-Tested Seamless GNSS Infrastructure in Metro CityACM MobiCom 2022 · H. Kim, J. Park, S. Park, J. Ryoo
- Which Uber is Mine? Identifying Target in Crowd of Objects with RF Analysis and AR Visual TagsACM MobiCom 2022 Poster · J. Park, H. Lim, J. Ryoo
- GNSS system supporting GNSS-denied areasKorean patent application No. 21-191864
Let's talk
Always happy to talk distributed-systems reliability, multi-cloud and test infrastructure.