We build the infrastructure between physical devices and the cloud: edge agents that never sleep, Kubernetes platforms that scale and the pipelines that turn raw signals into action.
Four disciplines, one continuous pipeline — from the sensor on a machine to the dashboard your team actually looks at.
Agents that plan, reason, and execute — from Telegram bots wired into n8n to LLM-powered assistants that query fleet status without ever touching raw credentials.
Go and C++ agents on Raspberry Pi and ESP32 stream host and sensor telemetry over MQTT, with JWT rotation handled in the background so fleets never go dark.
Multi-cluster fleets managed through zero-trust proxies, GitOps pipelines, and Helm engines built for one-click rollbacks, not late-night kubectl.
Prometheus, Loki, and InfluxDB stitched into dashboards that turn thermal sensors and vibration curves into decisions before failure, not after.
Working systems built and run by the lab — some in production, some still on the bench. Documentation is open where the system is live.
Distributed Edge Control Plane. A Go telemetry agent (gopsutil) monitors host and container health across bare-metal and cloud VMs, reporting into an async FastAPI backend with over-the-air JWT rotation and Discord/Slack alerting.
Self-hosted control plane for multi-cluster Kubernetes fleets. Kubeconfigs never reach the client — credentials are encrypted at rest with Fernet, sessions run on HttpOnly JWTs, and RBAC scoping maps team roles to real cluster permissions.
An end-to-end MLOps pipeline for industrial assets. Digital-twin simulators model ISO 10816 vibration standards; Random Forest models train on InfluxDB history and hot-load into AWS inference services for real-time diagnostics.
Multi-site telemetry ingestion on AWS IoT Core. An event-driven Lambda multiplexer evaluates thresholds and fires SNS alerts, with X.509 and mTLS enforced on every simulated device for strict topic isolation.
Tell us what's not talking to what — we'll figure out the flow.