Partner · AI / Fashion resale
Engineering process for an AI resale startup
Setting up Birl's first formal engineering process while contributing to the AI pipeline and desktop studio app.
- Client
- Birl Technology
- Role
- Tech Lead
- Year
- Jan – Apr 2026
- Link
- wearebirl.com ↗
- Stack
18Readiness issues found and remediated
~275Team commits a month across three repos
14Desktop app releases in one month
Architecture
How it fits together.
- ProcessTrunk-based branchingStaging gateTwo-layer review
- DeliveryGitHub ActionsTurborepoVercel previews
- PlatformCloud RunGKEOIDC authCloud Logging
- AIVertex AI image pipelineProduct embeddings
The brief
Birl builds an AI platform that helps brands manage second-hand inventory across eBay, Shopify and Depop. The team was growing and needed engineering process to match.
What we did
Mark led a team of four engineers across three repositories, while contributing directly to the code.
- Process. Documented Birl's first development process: dev, staging and production environments, trunk-based branching with a staging gate, AI-plus-human code review, CI/CD, QA sign-off and a release cadence. The founders signed off an 8-week rollout plan.
- Infrastructure. A three-environment GCP setup with Cloud Run, GKE namespaces and Vercel previews, and GitHub Actions pipelines with affected-only Turborepo builds.
- Production readiness. An audit of the resale API found 18 issues, from missing route authentication to webhook signature checks. We then implemented the fixes, including GCP OIDC authentication across all routes.
- AI pipeline. Migrated image processing from direct Gemini calls to Vertex AI, with clearer prompts, error handling and logging.
- Desktop app. Helped take the Electron studio app from prototype to production, with 14 releases in a month.
Have a project in mind?
Book a free 30-minute call to talk it through. No pressure, just a conversation about what you need.
Book a call