All case studiesTrust & safety · AI
Rizz AI Services
An AI-powered service integrating image analysis and inference workflows.
FastAPI
Python
- ONNX
- InsightFace
Screenshots

Problem
Trust and safety needed face inference without blocking realtime matchmaking or unclear model governance.
Face inference is license-sensitive and CPU-heavy—it cannot share a process with sub-second matchmaking without tail-latency risk and unclear model governance.
Solution
We isolated ai-service as Python 3.12 + FastAPI with explicit model directories and disabled auto-downloads, exposing narrow REST contracts the NestJS gender-verification module can timeout and degrade gracefully.
My role
FastAPI service implementation, ONNX/InsightFace inference paths, and NestJS integration contracts.
Architecture
- Isolated Python FastAPI inference service
- Operator-managed ONNX model directories
- NestJS HTTP client with timeouts, separate from matchmaking hot path
Tech stack
FastAPI
Python
- ONNX
- InsightFace
Client: Rizz Roulette
Impact
- Gender verification callable without blocking unrelated API traffic
- Clear separation between research-grade weights and production policy
- Room to add future inference modules without monolith growth
More case studies

