Building an Open Source RAG System for the United Nations Negotiations on Global Plastic Pollution
summary
An open-source RAG system supports the UN plastic-pollution treaty by ingesting chunked documents with multilingual sentence-transformer embeddings, storing vectors in Quadrant (1 GB free tier), driving retrieval via Haystack, and serving a Streamlit UI on Hugging Face Spaces. Retrieval uses mandatory filters (country, negotiation round), returns top-5 chunks, and limits LLM work to answers while programmatically attaching clickable references to avoid token bloat with long URLs; prototypes ran in real negotiations and continue to be used.
AI-written summary · about
chapters (10)
- 0:00Open source RAG system intro
- 5:13Technical architecture overview
- 6:37Document store selection
- 9:06AI framework choice
- 10:51LLM token limit optimization
- 14:44RAG performance levers
- 16:51Streamlit interface
- 18:37Code structure
- 19:40Public sector RAG challenges
- 22:33Q&A
related talks
35mThe Levels of RAG 🦜
59mRAG Against the Machine
34mImplementing an open-source RAG with OpenSearch
42mTake Your RAG To The Max!
26mAdvanced Document Parsing for AI Workflows
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
16mOpenRAG: An open-source stack for RAG
31mEasy RAG with LangChain4J and Docker