Not the one who signs the slides. The one who challenges me at 2 a.m.
Let's be clear: I don't use AI to write posts. I use it to build investment scanners, automate board packs, distil a degree into KPIs and keep myself honest about my biases. In everything I do, from consulting to my companies, from investing to studying to life, there's a co-founder who never gets tired and never flatters me. Here I tell you what I built and how deep I went. The recipe stays mine.
Where I put it to work, and how deep I went.
Automating every business process that runs through a computer: a huge time saver.
What I actually built. Where it touches clients or competitive edge, I leave out the details, on purpose.
A multi-step value-equity scanner that filters the investable universe in several passes, with backtesting discipline and explicit corrections for cognitive bias.
Real case, sensitive details omitted.
An SME scoring system for Italian companies: a predictive index and a succession-vulnerability triage, built on balance-sheet data.
Real case, sensitive details omitted.
Automation of a US SaaS's monthly board pack: from raw data to P&L, unit economics and KPIs with narrative, in minutes.
Real case, sensitive details omitted.
A full audit of an Italian luxury fashion brand's e-commerce infrastructure: it isolated a critical merchant-of-record lock-in and produced a 3-phase recommendation.
Real case, sensitive details omitted.
An app that consolidates portfolios from multiple brokers into a single view, with automatic enrichment by asset class, geography and sector.
Real case, sensitive details omitted.
Automation of the monthly expense report: from documents to filled-in forms, every month.
Real case, sensitive details omitted.
An operating system for decisions: a decision log, a cooling period and a 'council' of models that challenges my ideas before I execute them.
An economics degree distilled into an operating chart: 23 courses put through the sieve, 13 distilled, 8 business phases, 80 KPIs. It lives as a skill that questions me before a decision, and as a chart you print and hang.
The Vademecum turned into a tool: a web app with the 80 KPIs as checklists, 6 field calculators (cash conversion cycle, CLV/CPA, ROIC−WACC, DSCR, roll-up, share exchange), 18 real situations and search across 446 entries. At the end of a check it writes the verdict: where you are drifting off course.
Real case, sensitive details omitted.
The site you're reading and its blog: bespoke design, bilingual, with a pipeline that turns a conversation into a publish-ready SEO/GEO article.
A startup I co-founded: preserving grandmothers' recipes on video, with ingredient kits.
A framework for choosing sectors to acquire: defensible against AI and, at the same time, improvable with AI.
Real case, sensitive details omitted.
No hype. The tools, and the discipline I use them with.
Almost all the work runs through here: code, analysis, decisions. Not a chatbot, a workspace.
Connected to the APIs to give me every service I need.
They encapsulate my processes: due diligence, brand voice, council.
For big decisions, several AI advisors challenge me anonymously before I choose.
Where you need repeatable numbers, scoring is code, not 'model magic'. AI frames; it doesn't improvise.
I use it to build: investment scanners, board-pack and expense automations, a decision system, a site that publishes itself. Not to write posts, but to do the work and decide better.
No. It replaces the low-value hours, not the judgment. It amplifies someone who already knows what to look for; on its own it doesn't know what matters.
The scoring lives in deterministic Python, not in the model: repeatable, verifiable numbers. AI is there to frame and to keep me honest about bias, not to hand me a magic formula.
Mostly Claude Code, with Opus and extended thinking for hard problems, a library of bespoke skills, a council of models for decisions, and Python where you need solid numbers.
Yes, and sometimes the honest answer is no. I built a predictive SME scoring system and stopped it when the data didn't hold. That's value too.
Two reasons: client confidentiality and competitive edge. I'll tell you what I built and how far I got. Not the recipe that makes it defensible. The one exception is the Vademecum: that one you get in full, printable chart included — it's my degree distilled, not a client's competitive edge.
If it's worth my time, write me. If I can help, I'll answer. If I can't, I'll tell you.
Let's talk →