Turn Bilibili favorites into a personal RAG knowledge base

Name: Turn Bilibili favorites into a personal RAG knowledge base
Availability: InStock
Author: via2026

by via2026·Feb 22, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidSolve My ProblemNiche Gem

Bilibili-specific RAG pipeline with fallback ASR for inaccessible audio URLs.

Strengths

•Bilibili auth + audio extraction with 403-fallback handling is region-specific and clever — solves real platform friction.
•End-to-end retrieval: sync → ASR → embeddings → chat with source attribution; thoughtful pipeline design.
•Open source with diagnostic scripts (debug_asr, diagnose_rag) shows real ops thinking; cost warnings baked in.

Weaknesses

•'Chat with your video collection' is a solved pattern (e.g., NotebookLM, Glasp, local RAG tools); main differentiation is Bilibili-specific.
•Heavy dependence on DashScope (Alibaba) API for ASR and embeddings; costs unclear and tied to third-party SaaS.

Post Description

I built this to solve a personal problem: I save many long-form Bilibili videos (talks/courses), but cannot retrieve key ideas later.

Pipeline: - Bilibili auth + favorites sync - Audio extraction + ASR fallback (handles inaccessible audio URLs) - Chunking + embeddings + ChromaDB - RAG chat UI with source links

Stack: FastAPI, LangChain, ChromaDB, Next.js, SQLite.

I’d love feedback on: 1) retrieval quality tradeoffs 2) better indexing strategy for long videos 3) cost control for ASR + embeddings