About & contact

Roland Saïkali

Senior Engineer · Python · Backend · Data

Backend Python engineer, 25 years in prod. APIs, data models, pipelines, large-scale time series, at PagesJaunes, Airbus, EDF and Deutsche Telekom.

I build complete data chains: from raw ingestion to a usable model, reliability first. Simple, effective code that runs on its own for years, no 3 a.m. wake-ups.

Energy and IoT are my personal ground: LinkyStat ran for seven years, Home Assistant took over, this site is served from my homelab. That's where I test ideas, but the craft stays backend and data.

Roland Saïkali
Experience

Building Resilient Systems at Scale

Before building data infrastructure in my own lab, I designed critical backend systems for major industrial accounts.

Airbus via Sogeti / Capgemini

Engineered the Aerocity Data Management system ensuring the import and end-to-end traceability of aerodynamic calculations and complex 3D aircraft meshes (millions of nodes). A critical component to make simulations reliable and replace costly physical wind tunnel tests.

PagesJaunes via Sogeti / Capgemini

Transitioned from search engine development to infrastructure at the dawn of the DevOps movement. Fully automated provisioning and system configuration (Python, Fabric) to manage and scale large platforms, including Seety, hosting over 100,000 websites.

Deutsche Telekom via Marlysys

Developed DCSO, an Infrastructure-as-Code tool (YAML, OpenStack Taskflow) engineered to automate the TeraStream project (a disruptive 50x cloud-native network). Guided the client through modeling their datacenter topologies using our platform.

Method

Working with AI

Three stances today: the one who will never touch it, the one who ships without reading the code, and everyone else. I'm everyone else.

AI doesn't do my job. It pays the tax: the code that's necessary but decision-free, schemas, endpoints, plumbing. The time it gives back goes where decisions are made, among others:

Architecture. Why five services instead of a monolith. What breaks if the MQTT broker goes down, what degrades first, what rebuilds without loss.

Security. What's exposed and what must not be. Who can read what, where secrets live, what leaves my network, or doesn't.

Monitoring. Not "does it work", but "how will I know it stopped working", before the user does.

The user. The question there's never time for, yet the one that matters most: what's missing to help them in their daily life?

The list doesn't stop there: testing, data, costs, debt, everything that makes a system age well. Four examples are enough to show where the time goes.

Solo, all these hats are on my head. On a team, they're shared: the architect calls it, product owns the what, security audits. What doesn't change is the engineer who shows up with their questions already asked.

NILM is the proof: I'm not an ML specialist, I'm an engineer. Paired with AI, I cleared the ground, and twenty-five years of engineering discipline make the result trustworthy: clean data, thoughtful architecture, honest metrics published, code that holds in production.

Writing code was never the job. The job is knowing what you're building, how it holds, how it ages, and asking the questions before they get expensive.

Curiosity opens the doors; the craft stays in control.