Hello World, I'm cvag

Personal project · 2026

github.com/ChrisVagg/redfish-rollout-tool

Firmware rollout for server BMCs over the standard Redfish API: a fleet poller, a pre-flighted rollout in a canary and gated waves, and a lab of ten emulated OpenBMC GB200 NVL BMCs where the rollout really flashes firmware. It is built on the latest DMTF schemas with no vendor code, reads production ASUS, Dell, HPE, Supermicro and Gigabyte BMCs, and writes only to the lab. Beside them run the services a rollout needs: NetBox as the inventory's source of truth, a SeaweedFS image store that admits only packages whose signatures verify, an nginx HTTPS image cache in front of it that every update takes its image from, by push or pull, PostgreSQL holding a lock per host and every event of every run, and Prometheus, Loki and Grafana, which watch every BMC, keep its own logs and events and follow every rollout live. It is split as production would be: a control plane with the inventory, the store, the rollout state and the monitoring, and a site with its exporter, a Prometheus agent and the image cache, which only connects out.

• +11

Personal project · 2026

ProblemA portfolio needed to stay easy to edit as plain text, and deploy without hand-run steps.

A personal portfolio where every section is a Markdown file with front matter, and a single Node.js script turns the folder into a static page: projects, career, technologies and diagrams, with no framework and one dependency. The first version shipped as container images through GitHub Container Registry to a virtual machine behind Caddy; the current version is built and hosted by Cloudflare Pages.

• +5

Coolblock · Synapsecom Telecoms S.A. · 2023 — Present

ProblemNo single stack took a unit from thermal design to remote operation, on customer sites the company does not control.

The whole immersion cooling software solution, end to end — the parent of Coolblock Panel and Coolblock Portal. Every deployed unit runs its own Coolblock Panel, which controls the industrial system on site, and the Coolblock Portal watches and operates the whole fleet from one place.

Coolblock · Synapsecom Telecoms S.A. · 2023 — Present

Operates the whole current-model fleet of 12 deployed systems from one place, with zero downtime since launch

ProblemUnits sit on customer networks with no inbound access and nobody watching. There was no one place where the team could monitor, alert on and operate the whole fleet, and each customer only their own systems.

The fleet platform for the company's production immersion cooling systems: one place to monitor and operate every deployed unit, and a customer portal where each customer reaches only their own systems. Each system in the field runs its own Coolblock Panel, and the Portal reaches all of them across a zero-trust overlay network, while role & relationship-based access control scopes every user to the units assigned to them.

• +22

Coolblock · Synapsecom Telecoms S.A. · 2023 — Present

Deployed on 12 systems, every current-model unit

ProblemEvery model and PLC needs local control on sites nobody can easily reach, and with no one watching, each unit must raise its own alerts.

The on-site software solution for a single industrial system, running on the embedded industrial PC. The same containerised stack ships to every system in the fleet and is provisioned per installation by model and serial number, so one build serves the whole product line while each appliance drives exactly one machine. It is the field half of Coolblock Portal.

• +23

Coolblock · Synapsecom Telecoms S.A. · 2026 — Present

Monitors 20 systems, legacy and current models alike

ProblemLegacy units have no embedded PC to run the Panel, so their readings never left site, and every unit needed one lossless cloud path that spares the PLCs.

The telemetry path from the controllers of every immersion cooling unit into a cloud time-series store: a gateway reads the controllers and publishes their readings over MQTT, a broker carries them, and a subscriber transforms the raw payloads and writes them to the database. It picks up where Immersion Cooling IoT leaves off — the controllers that layer configures are the ones read here.

• +4

Coolblock · Synapsecom Telecoms S.A. · 2023 — Present

ProblemEvery unit ships with the same LAN addresses, yet engineers must reach any unit's devices remotely, without a site visit or a public port.

The IoT and network layer of an immersion cooling unit: the sensors calibrated and configured, the controllers that carry their readings, the fieldbus between controller and PLC, and the LAN everything sits on. It is the layer between the plant designed in Immersion Cooling System R&D and the software of Coolblock Panel — the path a physical reading takes before any application ever sees it.

Coolblock · Synapsecom Telecoms S.A. · 2026 — Present

Cools 200 kW combined heat on one loop

ProblemTanks sharing one dry cooler would fight over it, and without a single controller the water side stays passive, with nothing holding its temperature.

Automation for one cooling loop feeding several industrial systems in parallel: the pumps that pull water from the buffer, the dry cooler fan that rejects the heat, the logic that decides how hard each has to run, and the path from the plant's own OPC UA address space out to the company's metrics and dashboards. It applies the thermal design of Immersion Cooling System R&D to a multi-system installation, wired and networked the way Immersion Cooling IoT lays out a single one.

• +9

Coolblock · Synapsecom Telecoms S.A. · 2023 — Present

All models run at a PUE under 1.2

ProblemAir cooling tops out at a PUE of about 1.5 and cannot keep up with cooling dense HPC clusters.

The process side of the immersion cooling line, where everything serves the PUE target: the heat and flow balance that fixes where the system has to run, the component selection that follows from it, the PLC routines that hold it there, and the instrumentation it is all measured on. It is what Coolblock Panel reads and writes on site.

Synapsecom Telecoms S.A. · 2023 — Present

Monitors 2 data centres

ProblemGenerators, UPS, cooling, fire detection and PLCs each speak a different protocol, so the DC engineer had no single page to monitor, alert on and operate them.

Real-time monitoring and control for critical data centre facility systems: every device is polled over its own protocol, normalised into one time-series store, and visualised and alerted on in one place, from Uniflair precision cooling to VESDA fire detection.

• +4

Synapsecom Telecoms S.A. · 2023 — Present

Serves 60 tenants across 100+ racks, billed automatically every month

ProblemTenants are billed for energy, but only some racks have a PDU and meters were read by hand, so bills could not be traced or automated.

An end-to-end platform for collecting, storing and billing electrical consumption per data centre tenant. Racks fitted with a power distribution unit are polled over SNMP; racks without one are metered directly by current transformers measuring each phase. A scheduled Python job reads both, so every rack is accounted for regardless of how it is instrumented.

Nautobot is the source of truth for tenants, racks and infrastructure ownership. The collection logic lives in GitLab and is synchronised with Nautobot, so the job is scheduled, its changes are reviewable and its reports are structured. Measurements are written to InfluxDB and presented through Grafana.

• +6

Synapsecom Telecoms S.A. · 2023 — Present

Monitors 5 clusters running the company's VPS service

ProblemSubmerged servers run workloads that must not be touched, yet their temperatures, power and health still had to be measured. OS agents go silent when the OS is down and IPMI differs by vendor, so collection had to rely on the BMC alone, through one standard API.

Health, thermal and power telemetry for the cluster servers running inside the immersion cooling system, read out of band across Supermicro, ASUS, Gigabyte and Dell hardware. Each server's baseboard management controller is polled over the Redfish API, so readings come from the hardware itself and keep arriving whether the OS is up or down.

• +2

PD Neurotechnology · 2021

Used with over 100 patients

ProblemParkinson's symptoms were judged only by eye during a clinic visit, with no objective measure from the patient's daily life.

Symptom scoring for Parkinson's disease, built on signal analysis. A custom wearable records the patient's movement during guided tests and daily life, an Android app receives the recording, and the algorithm inside it turns raw motion into a score per symptom that is pushed to the cloud platform.

• +1

University of Patras · Diploma thesis · 2021 — 2022

ProblemPatients could not choose which doctor reads their vital signs, or take that access back.

A health monitoring and management platform where patients own their vital signs and grant or revoke a doctor's access on-chain. Conceived during the COVID-19 pandemic, it pairs a custom sensor board with two Android clients and an Ethereum contract that holds the authorisation logic.

• +5