React NativeExpo.NET 8PythonAPI

AI training system for dating app

A live system that collects and rates AI conversations to train a personality model.

Training an AI on personality requires a specific type of data that is not available for download. The solution was to build the data collection as a separate product. The system allows users to have a conversation with an AI persona and rate each response in real time. The feedback: passed, failed, or flagged, is fed directly into the training pipeline. Under the hood: a React Native interface, a .NET 8 API with an isolated database schema separate from the main app, and a Python FastAPI AI layer with Claude Haiku and Ollama as a local fallback. Prompt caching keeps cost and latency down. The design decisions were as important as the code: feedback blocks input (otherwise people skip it), all data is saved and filtered on export (not on collection), and the MVP schema is completely decoupled from the production database. The system is live and actively collecting training data.

Tekniker & verktyg

React NativeExpo.NET 8PythonAPIPostgreSQLpgvectorClaude HaikuDockerNext.jsTypescript

AI training system for dating app — screenshot 1