Case studies

Systems built, deployed, and running

I do not name my clients and I do not publish their figures without written consent. What I describe here is the architecture and the design decisions, that is, what lets you judge the work.

Some of these systems were built for clients, others for my own operations. I say which every time, because an internal system and a delivered one do not prove the same thing.

Client · IT consulting firm

Automated mission onboarding

When a mission is signed in the CRM, a chain triggers: creating the document tree, initialising the tracking record, updating the management base and notifying the right team. The hard part was not the chain but its reliability: the trigger is idempotent, so a webhook replayed by the platform never creates a second record, a classic failure of this kind of integration.

HubSpotn8nAirtableGoogle DriveSoftrTelegram
Client · Independent professional

Gmail inbox cleaned, then automated

A mailbox with tens of thousands of messages, no longer usable. Audit of volumes and senders, a folder taxonomy matching the real activity, backlog classification, then sorting rules. The decisive architectural choice: the bulk processing was done with my tooling, but the lasting system is a script attached to the client's own mailbox, running on its own every night. No subscription, no dependency on my infrastructure after delivery.

GmailApps Scriptn8n
Product · White label

Automated overdue invoice chasing

A chasing system designed to be deployed for several clients under their own brand: detection of overdue payments, a graduated reminder sequence, and immediate exit from the cycle once payment is seen. The exit logic is the sensitive part: chasing a client who has already paid costs more than the invoice.

n8nAirtablee-mail
System · Monitoring

Automated public tender monitoring

Watching tender notices, filtering on precise criteria (sector, amount, area, keywords) and alerting only on genuinely relevant ones. The challenge with this kind of system is not catching everything: it is not producing a feed nobody reads after three weeks.

n8nscrapingalerts
Internal · Conversational agent

Editorial assistant driven from a chat app

An assistant you write, revise and publish with, by chatting from an instant-messaging app. Every version is kept, you can roll back, retrieve an old post, and an editorial context memory grows over time. A rule written into the architecture: nothing is published without explicit approval, with protection against double-publication on technical replays.

Telegramn8nPostgreSQLLLM
Internal · Monitoring

Fleet monitoring for a site network

A dashboard watching the availability of a set of sites and the presence of expected technical elements in their pages, alerting on any deviation. It is also what lets me know a delivered system is still running, rather than hearing it from the client.

Node.jsExpressmonitoring
First conversation

A repetitive process eating your time?

Describe it in a few lines by e-mail: which process, how many times a week, how many people it burdens. I reply with a first read on what can be automated, and what cannot.

Reply within 24 business hours · No commitment · Confidential