Francesco
Marzano

Founder of LiftyUP.
Computer engineer.

Lives and works in Rome: Rome

My goal is to use AI to give people their time back, the most important thing we have. I do it by building products that bring real value to the people who use them every day.

Book a 30-minute callFollow me on LinkedIn
Francesco Marzano smiling, with glasses and a beard, in a checked shirt
Fig. 1, the founder in his natural habitat

I run LiftyUP

An AI-native software studio. We work on three fronts with one goal: giving hours back to people who spend them copying data and documents by hand.

A product that reads the client documents that don't go through the Italian e-invoicing system and prepares them for bookkeeping. The firm checks and approves.

Try the demo

AI agents, document extraction, knowledge bases on your own documents and ongoing development. Every project starts from an agreed goal.

See the services

20 tools built on official data and rules, from tax codes to severance pay. No sign-up.

Open the tools
Client documentsReady to book
Supplier
VAT number
Taxable amount
VAT to self-assess

Foreign invoices, receipts, expense notes. Every extracted figure is checked before it reaches the firm's accounting software.

The CV, with notes in the margin

On the left, what you'll find on LinkedIn. On the right, in red, what doesn't fit on a CV.

  1. 2017-2019

    DtoK Lab

    Full stack developer. I built a social data analytics platform with sentiment analysis and geographic analysis.

    my first code in production. Please don't ask me to reread it.

  2. 2019-2021

    NTT DATA

    Software developer consultant. I led three frontend developers on Angular applications, with a Spring Boot backend and Keycloak single sign-on.

    this is where I learned how long a meeting can last.

  3. 2021-2022

    Fides Consulting

    Full stack engineer on a corporate banking platform for a large Italian bank: Angular, Java and Spring Boot, strict security requirements.

    explaining software to the people who use it is half the job.

  4. 2022-2024

    Landytech

    Frontend engineer at a London fintech scaleup: React interfaces for investment reporting, code reviews and mentoring junior developers.

    standups in English, coffee strictly Italian.

  5. 14 June 2024

    Founding LiftyUP

    AI-native software studio: agents, automations and custom software for SMEs and professional firms.

    the day I took the leap.

  6. 2025

    Master in Product Management

    Product Heroes. To decide better what to build, before building it.

    perfect code nobody uses is just an exercise.

How I use AI

The model writes the code. My job is to build the harness around it, meaning the environment it works in: written rules, agents with a precise task and, above all, tests that decide what can go live.

harness
  • CLAUDE.md
  • Domain skills
  • Specialised agents
  • Hooks and gates
  • Progress log
  1. Plan (Analyst agent · read-only): A read-only agent studies the problem on real data. I approve the plan before a single line is written. plan approved
  2. Red test (Orchestrator): First I write the test and watch it fail, for the right reason. ✗ 1 test failing, as expected
  3. Code (Developer agent): A specialised agent, with the domain rules already loaded, writes code until the test turns green. ✓ the test passes
  4. Review (Reviewer agent · read-only): A second agent, with a clean context and no write access, looks for defects. Two rounds at most. independent review · round 1 of 2
  5. Gate (Deterministic script): Lint, types, the whole suite and a byte-by-byte comparison with the reference files. The first red stops everything. lint ✓, types ✓, suite ✓, byte by byte ✓
  6. Live (Automated end-to-end tests): We release and test again: on the deploy preview, then on the live site. deploy preview ✓, production ✓
  7. Lesson (CLAUDE.md): Every error we find becomes a written rule in CLAUDE.md, so it doesn't happen twice. +1 rule in CLAUDE.md

Every error we find goes back into the harness as a new rule.

How I test

  • Test firstNo logic without a test that has watched it fail first.
  • No blind trustThe model proposes, deterministic checks decide. A confidence score never makes a decision.
  • Byte by byteFiles for accounting software are compared with a verified reference. Almost identical means wrong.
  • On the real productAfter every release of the site, end-to-end tests in production: Italian and English, desktop and mobile.

The engine that reads documents for accounting firms has more than 10,000 automated tests. The method follows Anthropic's guidance on Claude Code best practices and on agent evaluation.

Education

Master's and bachelor's degree in Computer Engineering
University of Calabria
Master in Product Management
Product Heroes, 2025
Startup Builder
Startup Geeks, 2021
Scientific publication
BitTorrentSW, HPCS 2018 conference: reducing energy use in BitTorrent networks
Mentor
Startupbootcamp and Team4Idea

The pact

It applies to every project and comes before any quote. These are my values, in writing.

  1. Art. 1

    Clarity

    Before we start you know what I'll deliver, when, and what it costs. If you don't need something I'll tell you, even when the opposite would suit me better.

  2. Art. 2

    Professionalism

    Deadlines kept and code you can still read two years from now. If I get something wrong, I'm the first to say so.

  3. Art. 3

    Data decides

    Every solution is tried on your real data before it goes live. Opinions, mine included, come after the tests.

  4. Art. 4

    Flexibility

    Projects change while they're being built. When that happens we talk, recalibrate and move on, no drama.

  5. Art. 5

    Value first, invoice second

    A project only starts when it's clear how much time it will give back to the people using it. If it isn't clear, I don't sell it.

Francesco Marzano, founder of LiftyUP
my motto
You are youronly limit.

Write to me.
I'll answer.

Tell me about the process you'd like to get off your plate. In 30 minutes we'll work out whether it makes sense to automate it. If it doesn't, I'll tell you.

Book a 30-minute callMessage me on LinkedInRead the blog