React Native Mobile App

Leading a Native Code Developer Team in the Age of AI Coding

90% of the way through developing our frontend code, I became increasingly frustrated at the heaps of mismatched code dialects accumulating in the repository. Too many pull requests contained code with the personality of whichever AI coding assistant the developer used. Despite implementing clear policies and standardized guidelines early on, the bigger the codebase grew, the more diverse the LLM-assisted PRs became.

Instead of addressing this issue by limiting the use of AI to a single provider, or by outlawing it altogether, I took an approach that turned out to be massively beneficial for the team. I forked the repository and refactored the entire frontend to use a Feature-Sliced Design (FSD) architecture.

Not only did this produce positive outcomes in the universal interpretability of AI-assisted code, it also created a very intuitive architecture flow — enabling easier onboarding. We also found that it became smoother for multiple developers working on adjacent code simultaneously, and the hierarchical architecture that inherently demands strict adherence became somewhat self-regulating.

Full-stack architecture

The system spans a Postgres + Drizzle data layer, a Node.js API behind Nginx with Cloudflare WAF in front, real-time updates over Socket.IO and SSE, object storage via MinIO, and an internal Flask service proxying ML workloads in SpaCy, PyTorch, and TensorFlow. Tailscale links environments without exposing surface area.