app.py

import gradio as gr
import requests
import os
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_chroma import Chroma
from langchain_core.prompts import PromptTemplate
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough


# scrape docs from github using jina
def load_doc():
     urlsfile = open("./new_docs.txt")
     urls = urlsfile.readlines()
     base_path = "https://github.com/spacetelescope/notebook-ci-actions/tree/dev-actions-v2-pipeline/docs/"
     urls = [base_path+url.replace("\n","") for url in urls]
     pages = []
     for url in urls:
          url  = "https://r.jina.ai/" + url
          pages.append(requests.get(url).text)

     return pages
    


# embed to convert to tokens

embeddings = HuggingFaceEmbeddings(model_name="mixedbread-ai/mxbai-embed-large-v1")

# define the vector store and use it as retriever!

vectorstore = Chroma(
    collection_name="notebook_docs",
    embedding_function=embeddings,
    persist_directory="./",
)
vectorstore.add_texts(load_doc())
retriever = vectorstore.as_retriever()

# RAG prompt 
template = ("""
You are a GitHub Actions and Jupyter Notebook expert.

Your task is to answer the question **using only the information provided in the context**. 
If the context does not contain enough information, clearly state that and, if appropriate, 
briefly outline what additional details would be needed to give a complete answer.

Guidelines:
- Focus only on the question. Do not mention the context or that you’re using retrieved text.
- Provide step-by-step, technically accurate explanations and examples where relevant.
- Avoid speculation, guesses, or outdated practices. 
- Prefer modern, secure, and well-supported methods.
- Keep the answer concise but complete.

Context:
{context}

Question:
{question}

Answer:
""")
#template = ("""You are Github action and Jupyter Notebook expert.                                                                                  
#    Use the provided context to answer the question.                                                                                               
#    If you don't know the answer, say so. Explain your answer in detail.                                                                           
#    Do not discuss the context in your response; just provide the answer directly.                                                                 
#    Context: {context}                                                                                                                             
#    Question: {question}                                                                                                                           
#    Answer:""")

rag_prompt = PromptTemplate.from_template(template)

# define the LLM
llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash",temperature=0.1, google_api_key=os.environ.get('GOOGLE_API_KEY'))

rag_chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | rag_prompt
    | llm
    | StrOutputParser()
)


def rag_memory_stream(message, history):
    partial_text = ""
    for new_text in rag_chain.stream(message):
        partial_text += new_text
        yield partial_text


examples = ['What is the Unified Notebook CI/CD System?', 'How to migrate the old CI to the new CI?']
description = "Real-time AI App to Answer questions about the unified Notebook CI/CD system"
title = "Notebook CI/CD chatbot"

demo = gr.ChatInterface(fn=rag_memory_stream,
                        type="messages",
                        title=title,
                        description=description,
                        fill_height=True,
                        examples=examples,
                        theme="glass",
)

if __name__ == "__main__":
    demo.launch()