Production cloud infrastructure for signal processing, neural net classification, AI agents, and manufacturing systems — built to run unattended and scale when needed.
The Approach
Mobile and desktop apps handle the user experience. The cloud handles everything else — signal processing pipelines in R, neural net classification in Python/TensorFlow, data aggregation, statistical analysis, and AI agent orchestration.
This separation keeps apps fast and responsive while the compute-intensive work runs on properly sized infrastructure. GCP Compute Engine for persistent workloads, Docker containers on Cloud Run for stateless AI agents that scale to zero when idle.
The same approach scales from a single Debian VM running a manufacturing ERP to a multi-container architecture processing wearable sensor data from hundreds of participants.
The Stack
Production infrastructure that runs behind mobile apps, web dashboards, and manufacturing floors.
Debian-based VMs running production backends, databases, and long-running data processing jobs. Sized for the workload, not the hype.
Containerized microservices and AI agents. Cloud Run scales to zero when idle — you only pay for what you use. Perfect for LLM-powered chatbots and event-driven workflows.
Accelerometer, gyroscope, and magnetometer data processed server-side. Gait analysis, tremor detection, and biomechanical feature extraction at scale.
Time-series classification, speech emotion detection, predictive modeling. PhD research turned production code, running on cloud infrastructure with automated retraining.
Firestore real-time database, Firebase Auth, Cloud Storage, and Admin SDK. The glue between Flutter apps and cloud compute — with Python extraction pipelines to CSV.
CakePHP and MySQL on Debian Compute Engine. Multi-station manufacturing workflows with Android tablet interfaces at every station.
Case Studies
Real systems running real workloads — some for years.
Complete order and inventory management system for a hardwood flooring manufacturer. Debian Compute Engine running CakePHP/MySQL, with Android tablets at six stations: order entry, materials, packaging, production line, QA, and shipping. Real-time inventory tracking across the entire production floor.
Debian VM • CakePHP/MySQL • 6 Android stations • Still in production
GCP Compute Engine running R signal processing and Python/TensorFlow classification for wearable sensor data. Accelerometer and gyroscope streams from Heel2Toe devices processed server-side for gait analysis, with results pushed back to the Flutter companion app.
GCP Compute Engine • R • TensorFlow • Firebase integration • PLOS ONE published
Cloud backend for speech emotion detection in addiction recovery monitoring. Voice recordings captured on mobile, uploaded to cloud, processed through Python neural net classifiers, with emotional state scores returned to clinician dashboards in real-time.
Python/TensorFlow • Speech emotion detection • 623 clients • 7+ years in production
Dockerized deployment for HIV cognitive self-assessment at McGill. Python backend server, React frontend, and Traefik reverse proxy controlling URL access. Suite of cognition tests delivered through a containerized web application.
Docker • Python backend • React frontend • Traefik • McGill
Next-generation cognitive assessment platform for McGill. Flutter Web frontend with Firestore backend — lightweight, real-time, and scalable without the container overhead of the original BCAM deployment.
Flutter Web • Firestore • McGill
Containerized AI agents deployed on Cloud Run for client websites. LLM-powered chatbots trained on business data, running in Docker containers that scale to zero when idle — cost-effective infrastructure for always-on AI.
Docker • Cloud Run • Scale to zero • LLM integration
From a single VM to a multi-container architecture — let's talk about what you need running.
ted@appliedrd.com