Fábio Silva
AI as a solution,
not the goal.
Portfolio · Development
AI as a solution,
not the goal.
I'm Fábio Silva, a developer who starts with the challenge — the human, business, and constraint reality — and reaches for AI only when it's genuinely the right tool. Here's the proof.
Selected work
Challenges worth solving
The challenge
Recruiters pay to source candidates and then lose them — every new role starts from scratch, even when the right person is already sitting in a folder of CVs nobody can search.
The challenge
A solar developer's land team spends 2–3 days hand-cross-referencing environmental layers, grid capacity and slope for every candidate site — and most sites fail the first filter.
The challenge
Analysts lose most of their time to data preparation — fixing types, hunting duplicates, rebuilding the same Power Query flow for every new project — before the thinking can even start.
The challenge
Invoice data arrives as photos and PDFs in every layout imaginable — capturing it by hand is slow, and turning that pile into reliable, queryable spend indicators is a separate job entirely.
Who I am
I build software that starts with a real challenge and treats AI as one tool among many — reached for when it's the right answer, left out when it isn't.
My work tends to live where messy reality meets a clean interface: data that won't cooperate, processes nobody enjoys, decisions that need to be trustworthy. I care about restraint, legibility, and shipping things people actually keep using.
More about me →Ideas
Writing
Stop chasing the benchmark, start shipping
Every week a new model tops a new leaderboard, and every week someone rewrites their stack to chase it. Why the 'best AI model' debate is the wrong debate — and what actually separates teams that ship reliable AI from teams that don't.
ReadEurope lights up faintly on the AI map — and that should worry us more than the China panic
What the global distribution of AI computing power means for Europe — as a union, and for the companies that run on it. An interactive map of who owns the world's compute, and an argument for optionality over autarky.
ReadAgentic RAG: why your retrieval pipeline needs a brain
A retrieval pipeline that fetches but never reasons fails silently — a confidently wrong answer with no error thrown. Agentic RAG hands the LLM real agency over how grounding happens, and the engineering discipline that demands.
ReadGuides
Tutorials
The complete n8n self-hosting guide
A practical, production-minded walkthrough for self-hosting n8n — from npx and a single Docker container to Docker Compose behind a reverse proxy and full Kubernetes deployments across AWS, Azure, GCP, and OpenShift. Faithful to the official docs, with sources linked throughout.
ReadShipping AI in the enterprise: a builder's field guide
Most enterprise AI doesn't fail in the model — it fails in the last mile: integration, evals, rollout, people. A repeatable, phase-gated method for taking AI from idea to production, written from the project manager, AI engineer, and software developer seats at once.
Read