Low-Code versus Code-First RAG Helpdesk Chatbot with Flowise, Dify and LangGraph

Should a helpdesk chatbot be built with a low-code tool or written in code? This project builds the same retrieval-augmented chatbot three ways on a standard laptop - a Flowise chatflow (run over its REST API), a Dify workflow DSL (written; Dify needs Docker) and a LangGraph graph with a relevance gate - and measures the runnable ones on the same 60 questions from the public doc2dial Social Security help pages. Every build uses the same 798 chunks, embeddinggemma embeddings, top-4 retrieval, qwen2.5:1.5b on local Ollama and one prompt, so only the way of building differs.

NOTE: Answers come from public Social Security help pages (doc2dial, CC BY 3.0) and a small local model. They are drafts for a helpdesk person to check, not official advice; nothing is sent to a customer.

Category: Department of Computer Science and Engineering / Agent Development

Agent Development Project AG26 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.