Project N.O.M.A.D. inspires offline AI stacks

This is really an offline-infrastructure question dressed up as an AI question: the model is the easy part, the data pipeline, indexing, and power budget are the hard parts.

• –English Wikipedia alone is manageable, but adding images, maps, and multiple languages turns curation into a storage and maintenance project
• –OSM planet processing is the right instinct for offline navigation, but graph prep and edge/vertex extraction can dwarf the raw download size
• –For disaster or internet-shutdown use, small, curated libraries like Kiwix-style bundles will usually beat giant catch-all archives
• –Project N.O.M.A.D. is a strong reference stack because it already bundles local LLMs, offline maps, docs, and knowledge-base tooling into one install
• –The thread’s real takeaway: offline AI is viable, but only if you optimize for retrieval speed, portability, and energy use rather than model size