Built a RAG service on top of 10 classic Kyrgyz epics - ask anything, get answers in Kyrgyz. A small attempt to bring Central Asian literature to a wider audience.
Went with hybrid retrieval: OpenAI dense embeddings + BM25 sparse search, fused via RRF in Qdrant. Makes a real difference on Kyrgyz proper nouns, which a pure semantic search struggles with.
Frontend is intentionally styled after Kyrgyz shyrdak carpets - dark crimson, gold, geometric ornaments. 5-query guest limit via sessionStorage, no backend needed.
Biggest deploy headache: pywin32 sneaking into requirements.txt from a Windows machine. One line to fix, thirty minutes to find
Log in to leave a comment