EN FR
Independent engineering firm · Paris
Hardening, automation and recovery, for multi-cloud fleets and for high-volume communication platforms. One engineer named in the contract, a network of independent practitioners brought in when the work requires it, and the documentation that lets you take it back over.
We work at the operational layer: the one where enterprise platform rollouts succeed or stall. Multi-tenant provisioning, DNS and sender authentication architecture, deliverability engineering, identity federation, group-scale migrations. The work changes, the layer does not. We are brought in on systems nobody has mapped yet. The method does not change.
Four domains, one engineering discipline applied across all of them.
When one sender reputation carries every brand in the group, separating it without breaking it is an architecture problem, not a settings problem.
Policy says what should be true. The estate says what is. We work on the gap, at fleet scale and measurably.
Anything done by hand is not reproducible, and anything not reproducible cannot be restored. We put the estate into code before we touch it.
Enterprise AI rollouts rarely fail on the model. They fail on identity lifecycle, on agents holding more privilege than any human would be granted, and on a bill nobody owns.
We engage where communications infrastructure meets security and compliance requirements: retail, financial services and defense environments where resilience, access control and auditability matter as much as deliverability.
Our sweet spot is group structures running email at scale across multiple subsidiaries, where a single misconfigured sender has business-wide consequences. We are usually sponsored by IT Directors, Heads of Infrastructure or CISOs, and work hand in hand with marketing and CRM teams, from suppression list automation and bounce handling to campaign infrastructure, so that deliverability holds on both sides.
Whoever scopes the work delivers it. No substitution without your written approval.
Automation, version control and reproducibility, including on platforms usually configured by hand.
Documentation and runbooks are deliverables, not afterthoughts.
Air-gapped systems, defense-grade constraints, banking compliance.
Beyond the named engineer, we bring in selected independent practitioners where the work requires it. Named and approved with you before anyone starts.
Situations we are repeatedly called in on, described by pattern and order of magnitude.
A group grows by acquisition, or a subsidiary finally wants its own brand. Everything still ships from one account, one reputation, one pool of IPs, sometimes hundreds of millions of messages per month. Split it carelessly and every entity inherits the others' complaint history: password resets, order confirmations and invoices start landing in spam the week after cutover, and nobody can prove why. We map what actually sends before touching anything, then sequence the separation so reputation transfers instead of resetting.
Thousands of hosts, a datacentre contract with an end date, and an inventory nobody fully trusts. The migration itself is rarely the risk. The risk is the long tail: the few percent of machines nobody can identify, the dependencies that only surface at cutover, and the legacy environment that stays alive two more years because decommissioning was never anyone's objective. We treat the estate as code before we treat it as a migration: inventory reconciled against reality, configuration under version control, deployment automated, and high availability tested across on-premise, AWS and GCP.
A high-assurance environment borrows nothing from the internet: no update feed, no cloud console, no vendor telemetry. Every control has to be installed, configured and proven offline, and the audit trail has to survive the isolation. We deploy the stack as code: host and network intrusion detection, XDR/EDR, centralised identity, mandatory access control, log centralisation. All of it aligned with ANSSI and BSI, and reproducible without a network.
An expired card, a billing dispute, a closed account, and the production configuration is gone with it. Every hour costs transactional mail your customers are actively waiting for. You cannot simply recreate what was there: new IPs carry no reputation, authentication has to be re-proven to every receiver, and a rushed rebuild can cost more deliverability than the outage did. We reconstruct configuration from whatever survived (archives, exports, DNS history, message headers) and restore in the order that protects reputation. Back within a weekend.
It was configured by hand, by someone who has left, and it works. So nothing gets touched, upgrades are deferred, and one person's memory is the disaster recovery plan. We reverse the configuration into code, put it under version control, prove the rebuild in a throwaway environment, and hand back a system your team is allowed to change again.
We do not publish client names. The situations are real and so are the orders of magnitude. Named references available under NDA.
Platforms and tooling used in production engagements.
Scoping conversations are with the engineer who would do the work.
Direct