EventScope — Edge-to-Cloud Analytics Platform

Project Overview

Developed a fully distributed analytics platform for Johnson & Johnson Medical to analyze visitor engagement and booth interactions during the EADV 2025 medical congress. The system captures real-time behavioral metrics, such as dwell time, engagement rate, sentiment, and interactions using on-device computer vision and edge-based data processing.

Designed to operate entirely offline, EventScope delivers instant insights to event staff through a live dashboard while maintaining full GDPR compliance through on-device anonymization and face blurring. The platform provided J&J with a scalable, privacy-first analytics solution.

Technical Architecture & Implementation

Edge Layer: Each booth operated a local analytics node running on custom IoT hardware equipped with vision-based sensors. The nodes used lightweight TensorFlow Lite and MediaPipe models to perform emotion, gesture, and presence detection. All inference and data processing occurred locally, ensuring no raw visual data left the device.

Aggregation Layer: A central backend built on Rails 8 stack aggregated event streams, performing real-time data ingestion, deduplication, and transformation into visitor sessions, dwell times, and heatmaps. Solid Queue and Solid Cable provided background job orchestration and live WebSocket updates with sub-2s propagation latency.

Visualization Layer: A responsive Hotwire dashboard displayed real-time engagement metrics, including total visitors vs. engaged sessions, average dwell time, engagement rate, sentiment analysis, and interaction counters. All updates streamed live via database-backed WebSockets for maximum reliability in low-connectivity environments.

Technical Challenges & Solutions

Offline-First Design: Implemented a local-first event store with batch synchronization and exponential backoff to ensure zero data loss during network outages.

Edge ML Performance: Optimized concurrent inference of multiple ML models (emotion, gesture, OCR) to maintain 15–20 FPS throughput using adaptive sampling strategies and CPU-only execution.

Data Integrity: Designed idempotent ingestion endpoints and append-only raw event logs to guarantee deduplication and full traceability of derived analytics.

Zero-Touch Deployment: Used Ansible automation for configuration, provisioning, and network mesh orchestration, enabling new device deployments in under 20 minutes with no manual setup.

Results & Impact

• 10,000+ events/day processed with <100ms ingestion latency
• Zero data loss during offline operation
• 77% cost reduction vs. equivalent cloud infrastructure
• Enabled real-time booth performance insights and staffing adjustments

EventScope demonstrated that edge-first analytics architectures can achieve enterprise-grade performance and reliability while maintaining full privacy compliance and offline resilience — a crucial requirement for regulated medical environments.