Services
Services
Three kinds of work, each backed by a system I built and shipped to production.
Book a callAI customer support
An AI agent that drafts replies and account actions from each customer's real orders and history. Your team approves what it does until the results justify more autonomy.
Case study: A supervised AI support agent grounded in real customer data
Good fit when
- Most tickets need someone to look up an order, a subscription or a payment before they can be answered.
- You support several brands, markets or languages.
- Refunds, cancellations and account changes have to stay under human control.
What I build
- An agent connected to your help desk and order data that drafts replies in the customer's language.
- Rules in code, not in the prompt, for anything that touches money: who the customer is, what they own, what may be cancelled.
- A test set built from your real tickets that every prompt or model change has to pass before release.
Operations automation and internal tools
Automation for the work that happens between your systems, such as disputes, refunds, reporting and data syncs, and one internal dashboard where your teams run it.
Case study: One governed platform for a company's internal AI automation
Good fit when
- People copy data between tools, or someone rebuilds the same report every week.
- A no-code workflow proved the process is worth automating, and now fails on retries or volume.
- Each team has its own scripts and spreadsheets, and access is managed by hand.
What I build
- Workflows as code that retry safely and show what failed, instead of losing work without a trace.
- Reports and answers computed from your data warehouse, with the query behind each number.
- One internal dashboard with shared sign-in and role-based access, so the next tool starts with login, permissions and tables already done.
AI features in your product
Customer-facing AI, such as an in-app assistant, built to hold up with real users: quality you can measure, costs you can predict, and releases you can roll back.
Case study: One AI chat assistant service for several brand apps
Good fit when
- You want an assistant or another AI feature inside your product, built on your own data and rules.
- A prototype works in demos, but nobody can say whether the last change made it better or worse.
- Model cost, latency or provider outages matter at your volume.
What I build
- The feature itself, with your product data behind it and limits on what the model may say or do.
- Analytics over real conversations, so quality is measured rather than assumed.
- Cost and latency budgets, provider fallbacks, and a deploy that rolls back on its own when health checks fail.
How it works
- 01
A call to map one process: its data, its decisions and what an error costs.
- 02
A first piece that is useful on its own, agreed in writing with its risks.
- 03
Built to production standard: tests, evaluations, monitoring and a rollback path.
- 04
Handover: what to watch, and when the system can be trusted with more.
Pricing
Consulting is €80 per hour, in 60-minute blocks. Build work is quoted after the first call, for a defined first piece rather than an open-ended project.
Questions
How soon is something running?
The first piece usually runs within a few weeks. AI coding agents write much of the code, and tests, evaluations and a deploy that can roll back decide what ships, so the speed doesn't cost reliability.
Will AI replace my support team?
No, and it shouldn't be designed to. The agent does the lookups and first drafts; your team decides what it may do alone, one action at a time. Anything involving money or account changes can stay with a person for good.
What happens when the model is wrong?
It will be wrong sometimes, so the system is built for it. Every answer carries the evidence it used, tests catch regressions before release, unclear cases go to a person, and every decision is logged.
Is our customer data safe?
Data stays in your own infrastructure and accounts wherever possible. The model sees only the fields a task needs, access is scoped by role, and every action is logged. Model providers are chosen to match your data-processing terms.
Should we build this or buy a product?
Buy when the problem is common and the vendor's workflow matches yours. Build when the value depends on your own data, rules or process. Often it's both: a bought tool at the edge and a small system of your own where your data and decisions live.
