Turn Bilibili favorites into a personal RAG knowledge base
Bilibili-specific RAG pipeline with fallback ASR for inaccessible audio URLs.
Turbo-fast RAG for Frappe (v14) using TurboVec [https://pypi.org/project/turbovec/]
RAG for Frappe when LangChain and LlamaIndex already support custom integrations.
Frappe framework developers
LangChain · LlamaIndex · Haystack
Bilibili-specific RAG pipeline with fallback ASR for inaccessible audio URLs.
Consistent pseudonymization beats redaction when RAG embeddings must survive.
Predicts RAG benchmark transfer failure using vocabulary specificity—no embeddings needed.
Flow fields over embedding trajectories—TensorFlow Projector but with dynamics.
Using a single-file .pardus format with CREATE/INSERT/SELECT + SIMILARITY queries gives a very familiar developer UX for embedding storage. The combination of graph-based ANN, full transactions, thread-safety, and zero external dependencies is an uncommon and useful engineering combo for local-first AI work; it would win more attention with benchmark comparisons and richer ecosystem integrations (connectors/clients).
First public NRC regulatory embeddings dataset—37K chunks ready for ChromaDB and Pinecone.