I have a question about RAG AGENTS!
I have Supabase PostgreSQL with pgvector for embeddings, Dumpling AI for transcription and summaries, and a Telegram bot for chat with basic memory.
I want to know how to ingest raw text links or media into pgvector on the fly, tips for combining metadata with full text embeddings, the best approach to automatically rank and label documents in relation to a specific task such as identifying which tasks will become or are already AI aided, and strategies for maintaining long term memory in a Telegram based RAG bot.
If you have built or deployed a production grade RAG pipeline please share what worked or what failed and feel free to comment or send a direct message with serious insights. Thank you!!