Ceresio Labs
EN IT
Start a project

AI agency Custom AI for companies Milano · Lugano

AI that does real work for your company.

Ceresio Labs is an AI agency. We find the work in your company that an agent can take over, then design, build and run the system with Claude. Plain rules handle the routine; a person approves what matters.

Start a project Book a call
22,500+
job postings collected and classified by our own system since 21 August 2026
Every morning
an unattended run, logged, with an alert when something needs a look
Live on Amazon
a book translated with a multi-agent pipeline and revised by a person

Services

Custom AI for your company, and the systems behind it.

The first service is the one most companies need. The other four are the kinds of system we build, each one already running on our own products.

  • Custom AI for companies and offices Milano · Lugano

    We start from how your team works today, find the tasks that cost the most hours, and bring Claude into them: on your documents, in the tools you already use, with your people trained to work alongside it.

    Area served: Milano · Lugano · Lake Ceresio

    • Process auditWhere the hours go, and which tasks an agent can take over.
    • Internal assistants on your documentsAnswers from your own procedures, contracts and manuals.
    • Automation of repetitive workData entry, sorting, reports, follow-ups.
    • Claude in the tools you already useConnected to your systems, not another app to learn.
    • Team trainingPractical sessions on your real tasks.

  • Data collection and classification agents

    Daily collection from many sources, duplicates merged, every record sorted by your criteria. Rules decide the routine; Claude reads the cases rules can't settle.

    Running in Jobs Radar

  • Multi-agent content pipelines with human review

    Several Claude agents, each with one job: drafting, checking terms against a glossary, writing the notes. A person revises in depth and approves before anything is published.

    Running in Sottotesto

  • Operational automations on your server

    Scheduled jobs, backups before every risky write, push notifications when something needs attention. Standard Linux services, nothing exotic to maintain.

    Running in the Jobs Radar daily run

  • Fast Claude-native prototypes

    MCP servers that let Claude query your own data, small dashboards, command-line tools. A working prototype to try before deciding on a full build.

    Running in our MCP servers for job data and video transcripts

Work

Systems we built and run every day.

Jobs Radar

Data collection and classification

Problem

Tech job postings are spread over many portals. The same posting appears on different sites, and nothing shows the market as a whole, first across Europe and now in Ticino.

System

A collector runs every morning on a small server: it fetches new postings, merges duplicates across sources and tags role, stack and language requirements. Postings the rules can't settle are queued for review with Claude. A dashboard and an MCP server let a person, or Claude, query the result. Claude agents wrote the code and its tests.

Result

22,500+ European tech job postings collected and classified since 21 August 2026, with a run every morning. Ticino is the current focus.
Sources for Ticino: job-room.ch (SECO), jobs.ch, jobup.ch Internal tool, not public

Multi-agent content pipeline

Sottotesto

Classical texts in new Italian translations. Claude agents draft the translation and check every key term against a fixed glossary of 35 terms; a person revises in depth and approves. The copyright page says the translation was made with AI.

Vol. 01 live on Amazon: Seneca, La brevità della vita, paperback and Kindle.

See it on Amazon.it ↗

In the lab

What we're building next.

Each project carries its real status. No launch dates, no promises.

  • In design

    Learning engine

    A system for learning any subject with the «learn how to learn» method: it searches the course material, applies the method to the subject and runs recall sessions guided by Claude. On Mac and phone.

  • Prototype · in design

    Fluency coach

    A voice coach for fluency in English: a conversation every day with a voice agent that never interrupts, followed by targeted review.

  • In daily use · private

    Money management

    A personal finance app in daily use: accounts, budgets, savings funds and bank sync. An Angular app in production on our own server, built and maintained with Claude Code agents.

  • Open to pilot projects

    AI for SMEs

    A kit for bringing Claude into offices: an assistant on internal documents, automations and team training.

Process

From a process on paper to a system that runs.

  1. Discovery

    We look at the process as it runs today and find the part where an agent saves the most time.

  2. Spec

    A written spec: inputs, outputs, what rules decide, what Claude decides, what a person approves, and how we measure success.

  3. Build

    Claude agents write the code and the tests. A person checks the behaviour before anything ships.

  4. Run and monitor

    The system runs on a schedule, logs every run and sends an alert when something needs a look.

  • Rules first, Claude where it earns its place Anything a plain rule can decide, a rule decides. Claude takes the cases that need reading and judgement.
  • Every number has a source Data keeps the link to where it came from. Nothing estimated is presented as measured.
  • A person signs off Agents do the work; decisions about what ships stay with a human.

Founder

Alessandro Capozzi

Founder · AI consultant & engineer

A frontend developer who has spent years building software at an agency, Alessandro builds agent systems with Claude every day. Jobs Radar and Sottotesto are his: he designed them, built them with Claude agents and runs them.

Book a call

FAQ

Questions companies ask first.

Who do you work with?

Companies and offices around Milano, Lugano and Lake Ceresio. Much of the work can also be done remotely, in Italian or English.

Which AI do you use?

Claude, by Anthropic. Where a fixed rule is enough we use a rule: it is cheaper, faster and easier to check.

What happens to our data?

Before any build we agree in writing which data the system may see and where it runs: on your server or on ours. The system gets only the data the task needs.

How does a project start?

With an email describing the process. Then a call, then a written spec with scope and price. Nothing is built before you approve the spec.

Do you run the system after delivery?

If you want, yes: scheduled runs, monitoring and alerts, as we do for our own systems. Or we hand over the code with a guide your team can follow.

Contact

Tell us about the work you want to hand over.

A few lines are enough: what happens today, how often, and what a good result looks like. We reply with questions or a first proposal.

Start a project Book a call