
First-Aid Group Finder: Matching Unlinked Code Systems with Semantic Search
How semantic search with OpenAI embeddings matched two unlinked code systems: 3,257 Italian business codes and their first-aid risk groups, with a human in the loop.
home→projects
Some for clients, some for myself. Some for work, some for life(style). All of them explore problems people have. Can you find the one I've built for you?
I like building things that make our lives better: LLM pipelines that read and decide, automations that run — well — automatically, and personal projects that make us think about where we’re headed.

How semantic search with OpenAI embeddings matched two unlinked code systems: 3,257 Italian business codes and their first-aid risk groups, with a human in the loop.

After an AI diary that only cheered me on, I built one that shows me what I’d rather not see, links distant entries and offers hypotheses before conclusions.

How ChatGPT prompts, a custom WordPress theme and a Telegram bot turned a journaling practice into a diary that answered back (and what I learned).

How Python, a Jupyter notebook and Idealista data (via RapidAPI) mapped price-to-rent ratios across Las Palmas, and pointed me to the house I bought.

A sales chatbot for an advertising agency: it qualifies visitors, books calls with an expert, and a second AI step turns every conversation into an Excel lead list.

A tarot-reading cat built in early 2024 to learn the ChatGPT API: chained prompts, safety checks and bilingual readings, in plain PHP and JavaScript.

How I redesigned my website, practically untouched since 2014, in about ten days with Claude: one Project, different chats for different jobs, and a living doc to hold it all together.

A trading engine that ran several ML-driven strategies on Binance at once, with risk management built in. Paused while I re-architect it.

A 5-stage LLM pipeline that reads financial news every hour and reports the market effects, with evidence for each claim.

A live data pipeline and ML experiments to test whether Bitcoin’s order book can predict short-term price moves.

An LLM pipeline that reads my LinkedIn alerts, judges each company’s impact and scores every job against my profile, so 312 jobs become a shortlist of four.
contact
Need help with something? Looking for a professional who's reliable and believes in karma*?
Tell me about it here, or send me an email. I reply within two working days.
* which means value and integrity come first; reward follows.