import gradio as gr import pandas as pd import os import torch import zipfile import tempfile import shutil from bs4 import BeautifulSoup from typing import List, TypedDict from langchain_huggingface import HuggingFaceEmbeddings from langchain_community.vectorstores import Chroma from langchain_core.documents import Document from langchain_core.prompts import PromptTemplate, ChatPromptTemplate from langchain_core.output_parsers import StrOutputParser from langchain_core.runnables import RunnablePassthrough from langchain_nvidia_ai_endpoints import ChatNVIDIA from langchain_core.pydantic_v1 import BaseModel, Field from langchain_community.tools.tavily_search import TavilySearchResults from langgraph.graph import END, StateGraph, START import chromadb import io # Environment variables setup os.environ["TAVILY_API_KEY"] = "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" os.environ["LANGCHAIN_PROJECT"] = "RAG project" class GradeDocuments(BaseModel): """Binary score for relevance check on retrieved documents.""" binary_score: str = Field(description="Documents are relevant to the question, 'yes' or 'no'") class GraphState(TypedDict): """Represents the state of our graph.""" question: str generation: str decision: str documents: List[str] def process_documents(temp_dir): """Process documents from the extracted zip folder with enhanced error handling.""" d = {"chunk": [], "url": []} # Debug information print(f"Scanning directory: {temp_dir}") file_count = 0 processed_count = 0 error_count = 0 # Recursively traverse the directory for root, dirs, files in os.walk(temp_dir): for file_name in files: file_count += 1 file_path = os.path.join(root, file_name) print(f"Processing file: {file_path}") try: # Try different encodings encodings = ['utf-8', 'latin-1', 'cp1252'] content = None for encoding in encodings: try: with open(file_path, 'r', encoding=encoding) as stream: content = stream.read() break except UnicodeDecodeError: continue if content is None: print(f"Failed to read file {file_path} with any encoding") error_count += 1 continue soup = BeautifulSoup(content, "html.parser") title = soup.find("title") title_text = title.string.replace(" | Dataiku", "") if title else "No Title" main_content = soup.find("main") text_content = main_content.get_text(strip=True) if main_content else soup.get_text(strip=True) if not text_content.strip(): print(f"No content extracted from {file_path}") error_count += 1 continue full_content = f"{title_text}\n\n{text_content}" d["chunk"].append(full_content) d["url"].append("https://" + file_name.replace("=", "/")) processed_count += 1 print(f"Successfully processed {file_path}") except Exception as e: print(f"Error processing file {file_path}: {str(e)}") error_count += 1 continue print(f"\nProcessing Summary:") print(f"Total files found: {file_count}") print(f"Successfully processed: {processed_count}") print(f"Errors encountered: {error_count}") if not d["chunk"]: raise ValueError(f"No valid documents were processed. Processed {file_count} files with {error_count} errors.") return pd.DataFrame(d) def setup_rag_system(temp_dir): """Initialize the RAG system with the provided documents.""" # Initialize embedding model model_name = "dunzhang/stella_en_1.5B_v5" model_kwargs = {'trust_remote_code': 'True'} embedding_model = HuggingFaceEmbeddings( model_name=model_name, show_progress=True, model_kwargs=model_kwargs ) # Process documents df = process_documents(temp_dir) if df.empty: raise ValueError("No valid documents were processed") df["chunk_id"] = range(len(df)) # Create documents list list_of_documents = [ Document( page_content=record['chunk'], metadata={"source_url": record['url']} ) for record in df[['chunk', 'url']].to_dict(orient='records') ] # Setup vector store ids = [str(i) for i in df['chunk_id'].to_list()] client = chromadb.PersistentClient(path=tempfile.mkdtemp()) vector_store = Chroma( client=client, collection_name="rag-chroma", embedding_function=embedding_model, ) # Add documents in batches batch_size = 100 for i in range(0, len(list_of_documents), batch_size): end_idx = min(i + batch_size, len(list_of_documents)) vector_store.add_documents( documents=list_of_documents[i:end_idx], ids=ids[i:end_idx] ) return vector_store def create_workflow(vector_store): """Create the RAG workflow.""" retriever = vector_store.as_retriever(search_kwargs={"k": 7}) llm = ChatNVIDIA(model="meta/llama-3.3-70b-instruct", temperature=0) rag_prompt = PromptTemplate.from_template( """You are an assistant for responding to Request For Proposal documents for a bidder in the field of Data Science and Engineering. Use the following pieces of retrieved context to respond to the requests. If you don't know the answer, just say that you don't know. Provide detailed responses with specific examples and capabilities where possible. Question: {question} Context: {context} Answer:""" ) def format_docs(result): return "\n\n".join(doc.page_content for doc in result) rag_chain = ( {"context": retriever | format_docs, "question": RunnablePassthrough()} | rag_prompt | llm | StrOutputParser() ) return rag_chain def preprocess_csv(csv_file): """Preprocess the CSV file to ensure proper format.""" try: # First try reading as is df = pd.read_csv(csv_file.name, encoding='latin-1') # If there's only one column and no header if len(df.columns) == 1 and df.columns[0] != 'requirement': # Read again with no header and assign column name df = pd.read_csv(csv_file.name, encoding='latin-1', header=None, names=['requirement']) # If there's no 'requirement' column, assume first column is requirements if 'requirement' not in df.columns: df = df.rename(columns={df.columns[0]: 'requirement'}) return df except Exception as e: # If standard CSV reading fails, try reading as plain text try: with open(csv_file.name, 'r', encoding='latin-1') as f: requirements = f.read().strip().split('\n') return pd.DataFrame({'requirement': requirements}) except Exception as e2: raise ValueError(f"Could not process CSV file: {str(e2)}") def handle_upload(zip_file, csv_file, nvidia_api_key): """Handle file uploads and process requirements with enhanced error handling.""" try: # Set the NVIDIA API key from user input os.environ["NVIDIA_API_KEY"] = nvidia_api_key # Create temporary directory temp_dir = tempfile.mkdtemp() print(f"Created temporary directory: {temp_dir}") try: # Extract zip file print(f"Extracting ZIP file: {zip_file.name}") with zipfile.ZipFile(zip_file.name, 'r') as zip_ref: zip_ref.extractall(temp_dir) print(f"ZIP contents: {zip_ref.namelist()}") # Process documents print("Processing documents...") df = process_documents(temp_dir) print(f"Processed {len(df)} documents") # Preprocess and read requirements CSV print("Processing CSV file...") requirements_df = preprocess_csv(csv_file) print(f"Found {len(requirements_df)} requirements") # Setup RAG system print("Setting up RAG system...") vector_store = setup_rag_system(temp_dir) rag_chain = create_workflow(vector_store) # Process requirements results = [] for idx, req in enumerate(requirements_df['requirement'], 1): print(f"Processing requirement {idx}/{len(requirements_df)}") try: response = rag_chain.invoke(req) results.append({ 'requirement': req, 'response': response }) except Exception as e: error_msg = f"Error processing requirement: {str(e)}" print(error_msg) results.append({ 'requirement': req, 'response': error_msg }) return pd.DataFrame(results) finally: # Cleanup print(f"Cleaning up temporary directory: {temp_dir}") shutil.rmtree(temp_dir) except Exception as e: error_msg = f"Processing error: {str(e)}" print(error_msg) return pd.DataFrame([{'error': error_msg}]) def main(): """Main function to run the Gradio interface.""" iface = gr.Interface( fn=handle_upload, inputs=[ gr.File(label="Upload ZIP folder containing URLs", file_types=[".zip"]), gr.File(label="Upload Requirements CSV", file_types=[".csv", ".txt"]), gr.Textbox(label="Enter your NVIDIA API Key", type="password") ], outputs=gr.Dataframe(), title="RAG System for RFP Analysis", description="""<small>This agent helps you verify if a specific tool matches your project requirements by uploading your tool documentation and your CSV containing your requirements But first, visit NVIDIA LLaMA 3.3 70B and get your API key.</small> <br><br> Upload a ZIP folder containing URL documents and a CSV file with requirements to analyze. The CSV file should contain requirements either as a single column or with a 'requirement' column header. Enter your NVIDIA API key to use the service.""", examples=[], cache_examples=False ) iface.launch(share=True) if __name__ == "__main__": main()