Service · engineering
AI agents and business applications, built and shipped to production
The process runs, your teams supervise. Inbound documents read automatically, data entry eliminated, reporting produced without a thought, executed end to end in your Microsoft 365 tools and business apps, with human validation where it matters. When automation isn't enough, I build the application that's missing: business UI, back office, API. No stitching together no-code blocks when the job calls for real software.
This is for you if…
- A repetitive, high-volume process costs you hours every week: inbound documents, re-keying, reconciliations, follow-ups, reporting.
- The process is understood and quantified (ideally through the diagnostic), and the expected return is demonstrable.
- No off-the-shelf product covers your need, and you require an application built around your business rules.
- You want automations you own: hostable on your side, inspectable, with no black box and no lifelong proprietary platform.
- You require guardrails: traceability, confidence thresholds, human validation on sensitive decisions.
What I deliver
-
End-to-end workflows
Automatic triggering (email, file drop, schedule), AI processing, updates to your tools, notification. The whole process, not a fragment.
-
Custom business applications
When the process needs a real interface: data-entry and validation screens, back office, API, database. Built in .NET and Angular, tested, deployed, maintainable by a team other than mine.
-
Built-in guardrails
Full logging, confidence thresholds, a human validation queue for ambiguous cases. AI proposes; you keep control of what's sensitive.
-
Wired into your estate
Microsoft 365, SQL Server, your business apps, your network folders. No forced migration, no parallel platform to maintain.
-
Handover and supervision
Documentation, source code handed over, training for your teams, then monthly supervision if you want it: fixes and improvements as the models evolve.
The process
A first process in production within weeks, not an eighteen-month programme.
-
Scope
Scope, acceptance criteria and price agreed in writing before work starts. The case comes from your diagnostic roadmap, or from a pain point already quantified.
Deliverable: a signed scope document (what gets automated, what stays human).
-
Build
Iterative development on your real data, regular demos, business rules tuned with your teams. You watch the software take shape, you discover nothing at delivery.
Deliverable: the workflow or application in production on a first scope, supervised.
-
Extend and supervise
Volume ramp-up, extension to process variants, continuous monitoring. The system learns from your teams' corrections.
Deliverable: the full process at cruising speed + a monitoring dashboard.
Investment
A full-time data-entry role costs over €35,000 a year, errors not included. A process whose return was quantified in the diagnostic typically pays for itself within months. And you already have the numbers.
Investment
Fixed fee
Agreed in writing before work starts, depending on scope: from a first process on your existing M365 estate to a complete chain integrated into your IT system. Then monthly supervision, with no long-term commitment.
Frequently asked questions
Do we need the diagnostic before starting development?
No, if the process is already understood and the expected return is quantified: we scope it and start. The diagnostic is for when the question is still "where do we start", not "how do we automate this". Automating a cluttered process does not speed it up, it multiplies it: that is the only reason I ask the question before writing code.
How long before a first process runs in production?
Weeks for a first scope running in supervised production, not an eighteen-month programme. The initial scope is deliberately narrow: one process, one variant, real volume. Extending to the other variants comes next, once your teams have seen the system get something wrong and correct itself.
What happens when the AI gets it wrong?
It is designed for, not handled as an incident. Every run is logged, every decision carries a confidence threshold, and anything below that threshold goes to a human review queue instead of being written into your data. The AI proposes, your business rules dispose, and your teams keep control of the sensitive cases. The corrections they make feed back into the system.
What do we own at the end, and who can take over?
All of it: source code, workflows, documentation. Automations are built on n8n, which you can self-host and inspect, and applications in .NET and Angular, tested and delivered in your repository. No proprietary platform to pay for indefinitely, no black box. The bar I hold myself to is simple: a team other than mine must be able to take the system over.