A study space on AI agents.
Notes, experiments and open source code to understand how to build AI agents that actually reason. A personal laboratory open to anyone who wants to learn together.
Deep dives on AI agents, Knowledge Graphs, hybrid retrieval and emerging protocols. What I learn I write down, what I write down I publish.
Open source prototypes to test architectures, patterns and ideas. Runnable code, not just theory.
Technical books, articles and tutorials that make enterprise-grade techniques accessible to self-taught learners.
An open space: anyone who wants to read, discuss or contribute is welcome. No roles, no memberships.
Technical books for those who learn by doing
Amazon's trademark is used under license from Amazon.com, Inc. or its affiliates.
Articles, book previews, and open source projects — straight to your inbox.
Powered by Substack · One-click unsubscribe.
Open-source projects from the studio: Limen (landslide, flood and wildfire hazard across Italy), mcp-geo-server (GeoServer in natural language) and knowledge-graph (companion to the book "AI that Reasons"). More repositories will be published progressively.
Open AI map of landslide, flood and wildfire hazard across Italy: over 300,000 one-km² cells, every hour and with a 72-hour forecast. Open data only (ISPRA, CNR-IRPI, Copernicus, INGV, EFFIS, NASA FIRMS); a deterministic engine computes the numbers, and the per-region narratives are written with Claude. It complements, and does not replace, official civil-protection warnings.
Companion project of the book "AI that Reasons". Knowledge graph with ingestion pipelines, hybrid retrieval (vector + graph), web UI and multi-agent orchestration.
Intelligent MCP server to drive GeoServer in natural language (Microsoft Agent Framework): a chat + map UI resolves a request to the right layers and renders them with a thematic legend. Idempotent shapefile bootstrap into PostGIS, config-driven thematic SLD styles and a domain-agnostic catalog. Native arm64/amd64 Docker stack.
