AI-driven acne detection pipeline ingests mobile camera images through a streaming service, applies multi-stage ML models for lesion detection, classification, and severity scoring, then aggregates results into a unified patient score. Raw facial images flow through processing stages—face detection, bounding box annotation, binary confidence gating, and five-class lesion categorization—before weighted aggregation produces a final 0–100 severity score. This architecture demonstrates best practices for medical image processing: raw zone preservation, curated transformation with metadata tracking, and aggregated serving for low-latency API responses. Fork this diagram to customize model stages, add data quality gates, or integrate alternative classification frameworks for dermatology applications. The multi-image aggregation pattern ensures robust per-patient scoring despite single-image variability.