Aiki debuts local Wikipedia search, RAG

Aiki is more interesting as an educational build than a production retrieval stack: it strips RAG down to the parts you can understand, inspect, and modify without cloud services or heavyweight dependencies.

• –The custom TF-IDF retriever is the real hook here; it makes retrieval behavior transparent instead of hiding it behind a vector DB or framework defaults
• –Query expansion via Wikipedia links and redirects is a smart fit for this domain, since Wikipedia’s internal graph can recover a lot of synonym and alias coverage for free
• –The local-first setup keeps the barrier low for offline experimentation, especially for people who want to study retrieval quality without paying for APIs
• –Optional Ollama-based generation turns it into a full RAG loop, but the repo’s main value is in the search layer and pipeline clarity
• –This is a strong fit for tinkering, demos, and teaching, but likely not a drop-in answer for large-scale corpus search or production workloads