William Hung — Senior Backend Engineer

Creative team backend engineer. I build AI creative services, video generation pipelines, performance analytics, and reliability systems for the advertising platform.

Key Achievements:

• AI creative services, built from zero 140/140 crop safety · 4x CTR
  • Built a greenfield AI creative-decision microservice in FastAPI. It runs 4,300 lines across 3 repositories.
  • Its crop engine combines 4 computer-vision detectors. On a 140-sample benchmark it cut zero important areas.
  • Wrote the Go and libvips render pipeline underneath it.
  • Delivered dynamic video ads to 3 Asian markets. Click-through rate rose from 0.2% to 0.8%.
  • Drove the creative asset platform from wireframe to production, which let 4 frontend engineers work in parallel.

  Notes I wrote from this: Crop, Resize, Fill: Image Processing

• Serving reliability and efficiency at scale 10 req/2s → 200 concurrent
  • Re-architected screenshot capture. The old design shared one browser per pod, stalled near 10 requests every 2 seconds, and ran out of memory at 150 concurrent captures.
  • The new design runs standalone GPU pods that KEDA scales from 4 to 6. It handles 200 concurrent captures at a 99.5% success rate.
  • Raised the capture pass rate from 35% to 100%.
  • Replaced a process-local lock with a Redis distributed lock. That ended a batch-edit deadlock which needed a pod restart to recover.
  • Benchmarked 75 production creatives over 5,625 runs. GIF to WebP cuts file size 51% at an SSIM of 0.994.

  Notes I wrote from this: From 19 Goroutines to Exit 137Remote-Controlling Chrome in ContainersImage Format Benchmark: WebP WinsThread-Safe Login, the Hard Way

• Security ownership CVSS 9.1 · fixed in 6h
  • Led the response to a CVSS 9.1 Critical vulnerability. A GraphQL endpoint had run with no authentication since 2019 and exposed about 41,933 production records.
  • Shipped an IP-whitelist fix within 6 hours.
  • Audited about 1.4 million log entries and confirmed no data exfiltration.
  • Designed the AppsFlyer signed-click service, and moved Kafka producers from plaintext to SSL across 4 repositories.
  • Closed 4 CVEs in a Go upgrade. A network policy I raised myself closed a lateral-movement path.

  Notes I wrote from this: Seven Blind Spots Behind a JWT AuditPKI & TLS Behind a Migration TicketA Golang Upgrade's Chain Reaction

• Platform and delivery harness 10-command contract · 33 ingress gated
  • Built a reusable delivery harness for AI services. It covers Jenkins pipelines, pull-request governance with a hotfix guard, version-gated smoke tests, and Prometheus observability.
  • Rolled it across the backend fleet under one 10-command cicd.sh contract. The image service went from 4 to 10 commands, and the preview service went from 0 to 7.
  • Put 33 Kubernetes ingress templates across 12 apps behind an automatic review gate.
  • Wrote the first architecture decision record for that repository.
  • Moved services from virtual machines to Kubernetes, which cut infrastructure cost 35%.
• Data pipeline and freshness 47 → 19 min · staging 7-day sync
  • Moved the creative-performance pipeline from fixed schedules to storage events. Trigger latency fell from 47 to 19 minutes, and data freshness fell from hours to seconds.
  • Pipeline runs dropped 40% to 60%.
  • Built an Argo CronWorkflow that copies production PostgreSQL into staging and development every week. Staging had no sync at all before this.
  • Credentials come from a vault, and write guards reject any mutation on a restored row.

  Notes I wrote from this: Reaching a Remote DB From Your LaptopOne Operation, Many Engines

• AI-driven velocity, mine and the team's 11 → 75 SP/month
  • Raised my sprint velocity from 11 to 75 story points a month with an AI iteration workflow I built myself.
  • Turned that personal tooling into a governed plugin marketplace with a required spec, continuous-integration checks, and a one-command install.
  • Built an AI operations bot. It is a Go daemon that turns a Slack mention into a sandboxed Claude Code session.
  • Every tool permission returns to the Slack thread as an emoji approval, and no bypass flag exists. The daemon carries 63 tests.
  • It shipped 2 production fixes straight from Slack. One merged 1 hour and 54 minutes after the alert.

I built enterprise cloud security products, V1FS and V1CAM. They run across AWS, Azure, GCP, and Alibaba Cloud. My work covered cloud asset security and event-driven bucket protection.

Key Achievements:

• Designed the authentication system on OIDC with least-privilege rules. It tightened how customers manage access.
• Broke past the AWS CloudFormation resource limit with the Serverless Compose framework. Deployment time fell 66%, and the infrastructure stayed stable.
• Built the file security products V1FSS and the V1FS SDK. They scan files in the cloud and on-premises.
• Wrote multi-cloud code for AWS, Azure, GCP, and Alibaba Cloud in Go, Node.js, and Python.
• Set up the delivery pipeline with Jenkins, GitHub Actions, Docker, Kubernetes, and Terraform. That shortened the path from commit to release.
• Earned the AWS Solutions Architect Associate certification (SAA-C03).

Reference:

V1FS, V1CAM, OIDC, CloudFormation limit, Serverless Compose, V1FSS, V1FS SDK, SAA-C03

I designed and built a cloud learning platform for AWS and Azure. It included interactive training environments and assessments.

Key Achievements:

• Designed a cloud-native learning platform. It combined mind-map views with hands-on labs for certification practice.
• Presented a computer-vision solution at the company AI conference.

Reference:

eCloudValley Official, Conference Link

Key Achievements:

• Joined the 5G and 6G research group, and the blood-pressure patent project.

Reference:

Academia Sinica Official, Blood Pressure Patent

Key Achievements:

• Joined the 5G research team. We won first place in the national R&D competition.

Reference:

QCT Website, R&D 1st place award news