Gemini CLI orchestrates a RAG reference migration

Remigiusz Samborski used Gemini CLI with the Conductor extension to coordinate a RAG migration reference architecture. The work crossed Terraform infrastructure, Python services, a Next.js interface and data pipelines involving BigQuery and AlloyDB. His account links the project repository and its saved specifications and implementation plans.

Each feature became a track. Gemini inspected the codebase, asked questions and wrote a specification and plan. Samborski reviewed those documents before implementation began. The agent then iterated through tasks, writing failing tests, implementing changes, refactoring and committing the work. Project instructions required Terraform plans before applies.

At phase boundaries the agent ran verification, supplied manual checking instructions and recorded results in Git notes and the plan. A sandbox notifier let Samborski return when his input was needed. He also used Antigravity for later code and documentation refinements.

The reported result was a multi-phase reference project with an auditable development history, rather than proof of a production migration at a customer. Samborski reports about $30 in Vertex AI token costs across several days, using several Gemini models and substantial cached input. That figure describes his run and excludes a general cost guarantee. Human approval, manual checking and the saved checkpoints remained part of the workflow throughout.

Prompt
Suggested prompt (written by Agents In Use, not the author's)

Source: DEV Community · Remigiusz Samborski