Applied AI lab · Switzerland

AI systems, built and tested on real problems. Every experiment documented: what works, what breaks, what it costs.

I design AI agents and automated workflows, run them on real work, and publish the results so you can judge before you buy.

Avatar studio pipeline
  1. IdeaHumanA one-line brief for a look
  2. Scout agentWeekly scan of the catalogue for demand gaps
  3. Production agentsSpecialised agents, each owning one step
  4. ReviewHumanApprove, reject or send back
  5. Item libraryApproved items, ready to publish

Experiments

Each one starts from a real problem and ends in a write-up.

  • Status:Running

    Avatar studio

    A team of specialised AI agents turns a one-line idea into finished Roblox avatar looks. I review; the agents produce. A weekly scout flags gaps in the catalogue worth filling.

    Stack: Multi-agent production · Market scouting · Human review

    Write-up in progress

  • Status:In production

    Voice receptionist

    A bilingual phone agent for a specialist practice in Geneva. It books and cancels appointments, sends confirmations and summarises every call. Call data leaves the automation layer within 7 days.

    Stack: Vapi · ElevenLabs · Make · Google Calendar · Brevo

    Write-up in progress

  • Status:Planned

    Weak-signal scout

    Watches new listings on Chinese platforms for products and business models before they reach Western trend tools. One digest a week, one source to start.

    Stack: n8n · Python · Weekly digest

    Write-up in progress

How the Lab works

Four rules, applied to every experiment.

  1. Real problem

    Start from work someone already does by hand, not from a tool looking for a use.

  2. Smallest working system

    Build the least that runs end to end, then improve what the results point to.

  3. Measure

    Cost per run, failure modes and time saved, tracked from the first day.

  4. Publish

    Write it up, including what didn't work. The write-up is the deliverable.

Work with me

The same method, applied to your business.

  • Project build

    One defined system, scoped, built and handed over with its documentation.

    • Voice agents and automated workflows
    • Fixed scope, fixed timeline
    • Running costs estimated before you commit
  • Fractional AI lead

    Part-time ownership of your AI roadmap: what to build, what to buy, what to skip.

    • Opportunity audit across your team
    • Hands-on builds, not slideware
    • Monthly engagement

Scope and price are agreed on a first call.

Book a first call

About

I'm Benjamin Duband. I spent 18 years in marketing and digital at Swiss companies, including Logitech and Swissquote. I now build AI systems hands-on and publish what I learn.

Recepio Lab is where that work lives: real problems, working systems, honest write-ups.

  • 18 years Marketing and digital
  • Valais Switzerland
  • EN · FR Working languages