Skip to content
← Back to the desk
No. 06 — AI Play policy checker · 2025

App Store Reviewer

App Store Reviewer simulates a Google Play review before you submit. It parses an app's AndroidManifest.xml, checks its permissions against Play policies, scores how likely the app is to be rejected, and writes the kind of comments a real reviewer would — so problems surface before Google finds them.

View on GitHub ↗
App Store Reviewer banner

what I built

  • Classifies every Android permission in a manifest by risk level — HIGH, MEDIUM or LOW.
  • Cross-references permissions against Google Play policies (SMS, location, user data and more) to flag likely violations.
  • Calculates a rejection probability from the severity and number of violations found.
  • Grounds the model with RAG over embedded policy documents, then generates reviewer comments with a LoRA fine-tuned Llama 3.2 served through Ollama.
  • Full-stack: a Next.js dashboard for upload, detailed reports and history, over a Django REST API that orchestrates the pipeline.

built with

Next.jsDjangoLlama 3.2RAGOllama