RAG Chatbot Pipeline

# Dify RAG Chatbot — OSOP Portable Workflow
#
# Retrieval-augmented generation pipeline: receive a user question,
# retrieve relevant chunks from a vector store, rerank for relevance,
# generate a cited answer, and collect user feedback for fine-tuning.
#
# Build in Dify or validate: osop validate dify-rag-chatbot.osop.yaml

osop_version: "1.0"
id: "dify-rag-chatbot"
name: "RAG Chatbot Pipeline"
description: "Question → retrieve → rerank → generate answer with citations → feedback."
version: "1.0.0"
tags: [dify, rag, chatbot, vector-search, llm]

nodes:
  - id: "receive_question"
    type: "api"
    subtype: "webhook"
    name: "Receive Question"
    description: "User submits a question via chat interface or API."

  - id: "retrieve_docs"
    type: "db"
    name: "Vector Retrieval"
    description: "Embed the query and retrieve top-k chunks from the vector store."
    config:
      store: "pinecone"
      top_k: 20
      embedding_model: "text-embedding-3-small"

  - id: "rerank"
    type: "agent"
    subtype: "llm"
    name: "Rerank Results"
    description: "Cross-encoder reranking to select the most relevant 5 chunks."
    config:
      model: "cohere-rerank-v3"
      top_n: 5

  - id: "generate_answer"
    type: "agent"
    subtype: "llm"
    name: "Generate Answer"
    description: "Produce a grounded answer using retrieved context with inline citations."
    config:
      model: "gpt-4o"
      temperature: 0.3
      system_prompt: "Answer using ONLY the provided context. Cite sources as [1], [2], etc."

  - id: "format_citations"
    type: "cli"
    subtype: "script"
    name: "Format Citations"
    description: "Map inline citation markers to source document metadata."

  - id: "collect_feedback"
    type: "human"
    name: "User Feedback"
    description: "Thumbs up/down and optional comment for answer quality tracking."

  - id: "log_for_finetuning"
    type: "db"
    name: "Log to Training Set"
    description: "Store high-rated Q&A pairs for future model fine-tuning."

edges:
  - from: "receive_question"
    to: "retrieve_docs"
    mode: "sequential"
  - from: "retrieve_docs"
    to: "rerank"
    mode: "sequential"
  - from: "rerank"
    to: "generate_answer"
    mode: "sequential"
  - from: "generate_answer"
    to: "format_citations"
    mode: "sequential"
  - from: "format_citations"
    to: "collect_feedback"
    mode: "sequential"
  - from: "collect_feedback"
    to: "log_for_finetuning"
    mode: "conditional"
    when: "feedback == 'positive'"
    label: "Good answer — save for training"