Build Document RAG System with Kimi-K2, Gemini Embeddings & Qdrant

This advanced n8n workflow constructs a complete RAG pipeline for processing large documents, such as the UK Highway Code or educational materials. It imports documents via HTTP, extracts text, and processes each page individually by generating contextual summaries using the surrounding pages (previous and next) with the Kimi-K2 Instruct LLM from Moonshot AI. This approach ensures summaries capture broader context, improving retrieval accuracy over raw text embeddings.

The workflow then converts these summaries into high-quality embeddings using Google's Gemini-embedding-001 model via a custom HTTP request (due to n8n compatibility). Finally, the embeddings are upserted into a Qdrant vector database collection, ready for querying by agents, MCP servers, or other workflows. Integration with Featherless.ai for inference allows unlimited token usage under a flat fee, making it cost-effective for high-volume summarization.

Benefits include reduced token costs for RAG prep, scalable handling of lengthy docs without truncation issues, and superior semantic search via contextual embeddings. Ideal for education (online courses, textbooks), compliance docs, manuals, or any knowledge base needing precise retrieval.

Use cases: Automate RAG for course platforms, legal/policy docs, technical guides; integrate with chatbots for contextual Q&A; preprocess corpora for AI agents.