gemini-review

To get a second opinion review for AI translation pipeline outputs, you can run a Gemini-based review script that independently evaluates ULT, UST, notes, issues, and alignment stages to catch errors a single model might miss.

Yes, you need the Gemini CLI installed in your environment to perform automated USFM reviews, as the script reads workspace guideline docs and outputs to generate findings.

You generate a findings report by running the Gemini review script for a chosen stage like ult, issues, ust, notes, alignment-ult, or alignment-ust, targeting a specific book and chapter.

Running a parallel Gemini review provides an independent quality check on pipeline outputs, computing structured findings that facilitate fast iterations and traceability during development or production runs.

When reviewing translation outputs, the findings reports are written as structured markdown files to the output/review/{BOOK}/{BOOK}-{CH}-{stage}-gemini.md path for traceability and subsequent fixes.