Skip to main content

Tso Chinese

Led Tso 3.0's technical direction and cross-functional team, then wrote the phased database migration plan the team adopted.

Role
Lead engineer, solution architecture
Scope
  • Solution architecture
  • Team leadership
  • Preview environments
  • Delivery operations
Tso Chinese website
A bowl of glazed chicken with broccoli and dried chilies
A preview environment per pull request

Labeling a pull request deploys it to its own namespace with a database, test data, RabbitMQ, and Jaeger tracing. GitHub Actions and Go handle the deploy and the cleanup, and the CTO, QA, and PM try the branch there before merge.

Use plus and minus to zoom, arrow keys to move around the diagram when zoomed, and Home to reset.

Contribution 04. Preview environments: a label on a pull request deploys the branch to its own namespace, so the CTO, QA, and PM could try it before merge.

Ordering, dispatch, and kitchen software for a dark-kitchen delivery business.

I developed and supported Tso 2.0 as the business grew from two to five restaurants, then designed Tso 3.0's architecture. It went live in September 2026.

Tso Chinese is a US restaurant brand that runs dark kitchens: an ordering site, an admin panel, a dispatch screen for delivery drivers, and kitchen tablets. Tso 3.0 is the redesign for 2.0's downtime releases and single-Postgres bottleneck.

Product
Dark-kitchen automation
Focus
Microservices, GraphQL, Team lead

What I did at Tso Chinese

  1. Developed and supported the Tso 2.0 ordering and kitchen platform, owning client communication, releases, monitoring, and incident response. Migrated payment processing from Stripe to Cybersource to cut processing costs.

  2. Reduced the database load from kitchen tablets that re-fetched on every change signal and hit Postgres in waves, first with query analysis and Redis, then by sending only what changed so only the affected tablet refreshed.

  3. Designed Tso 3.0 as solution architect around the CTO's two requirements: test changes in small, isolated pieces before they reach the kitchens, and grow the team without people stepping on each other. That meant Kubernetes, a database per service, RabbitMQ, and GraphQL gateways, so each service scales on its own, releases need no downtime, and one engineer can own payments without touching ordering.

  4. Built preview environments: label a pull request and it deploys to its own namespace with a database, test data, RabbitMQ, and Jaeger tracing, so the CTO, QA, and PM could try any branch before merge. GitHub Actions and Go handled deploy and cleanup, and the AWS calculator showed previews costing less than full environments.

  5. Led a cross-functional team of about 7 to 9 while staying hands-on, setting priorities with the CTO and coordinating delivery with the PM. Later, as a consultant, documented the migration in phases across off-hours windows, with cutover failure scenarios and mitigations, and presented it to the team.

A live commerce system where a bad architecture call would have hit the kitchens, drivers, and payments.

More client work

f6f7778