Neom99 commited on
Commit
6facec8
·
verified ·
1 Parent(s): e14307f

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +186 -0
app.py ADDED
@@ -0,0 +1,186 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import gradio as gr
3
+ from dotenv import load_dotenv
4
+
5
+ # --- RAG Libraries ---
6
+ from langchain.document_loaders import PyPDFLoader, TextLoader, Docx2txtLoader
7
+ from langchain.text_splitter import RecursiveCharacterTextSplitter
8
+ from langchain.embeddings import OpenAIEmbeddings
9
+ from langchain.vectorstores import FAISS
10
+ from langchain.chat_models import ChatOpenAI
11
+ from langchain.chains import RetrievalQA
12
+ from langchain.prompts import PromptTemplate
13
+ from langchain.schema import Document # Needed for creating custom documents from Excel
14
+
15
+ # For Excel file handling
16
+ import pandas as pd
17
+
18
+ # Load environment variables (for API keys when running locally)
19
+ # On Hugging Face Spaces, secrets are automatically loaded as environment variables.
20
+ load_dotenv()
21
+
22
+ # --- RAG Setup ---
23
+ # Initialize embeddings and LLM globally for efficiency.
24
+ # This assumes OPENAI_API_KEY is available as an environment variable.
25
+ rag_initialized = False
26
+ try:
27
+ embeddings = OpenAIEmbeddings()
28
+ llm = ChatOpenAI(temperature=0.7, model_name="gpt-3.5-turbo")
29
+ rag_initialized = True
30
+ print("OpenAI embeddings and LLM initialized successfully.")
31
+ except Exception as e:
32
+ print(f"Error initializing OpenAI components: {e}")
33
+ print("Please ensure OPENAI_API_KEY is correctly set in your environment or Hugging Face Space secrets.")
34
+ print("RAG functionality will not work without a valid OpenAI API key.")
35
+
36
+
37
+ def get_document_content(file_path, file_extension, filename):
38
+ """
39
+ Loads and returns text from various document types, including Excel, using Langchain loaders.
40
+ For Excel, it converts each row into a Langchain Document object.
41
+ """
42
+ print(f"Attempting to load document: {filename} with extension {file_extension}")
43
+ documents = []
44
+
45
+ if file_extension in ['.xlsx', '.xls']:
46
+ try:
47
+ df = pd.read_excel(file_path)
48
+ # Convert DataFrame rows to text chunks for RAG
49
+ for index, row in df.iterrows():
50
+ # Customize how you want each row to be represented as text for the LLM
51
+ # Example: F"Row {index+1}: {row.to_json()}" provides all column data as JSON.
52
+ # Or for more readable context, pick specific columns:
53
+ # doc_content = f"Row {index+1} from '{filename}': " + ", ".join([f"{col}: {val}" for col, val in row.items()])
54
+ doc_content = f"Row {index+1} from '{filename}': {row.to_json()}"
55
+
56
+ documents.append(Document(page_content=doc_content, metadata={"source": filename, "row_number": index+1}))
57
+ print(f"Excel file '{filename}' processed into {len(documents)} Langchain Documents.")
58
+ except Exception as e:
59
+ raise ValueError(f"Could not process Excel file '{filename}': {e}")
60
+ elif file_extension == '.pdf':
61
+ loader = PyPDFLoader(file_path)
62
+ documents = loader.load()
63
+ elif file_extension == '.txt':
64
+ loader = TextLoader(file_path)
65
+ documents = loader.load()
66
+ elif file_extension == '.docx':
67
+ loader = Docx2txtLoader(file_path)
68
+ documents = loader.load()
69
+ else:
70
+ raise ValueError(f"Unsupported file type: {file_extension}. Please upload .pdf, .txt, .docx, .xlsx, or .xls files.")
71
+
72
+ if not documents:
73
+ raise ValueError("Could not load any content from the document. It might be empty, corrupted, or an unrecognized format.")
74
+ return documents
75
+
76
+ def analyze_document_with_rag(file, query):
77
+ """
78
+ Core RAG function that processes the uploaded file and answers the query.
79
+ This function is directly exposed by Gradio.
80
+ """
81
+ if not rag_initialized:
82
+ return "RAG system not initialized. Please ensure your **OPENAI_API_KEY** is configured correctly in Hugging Face Space secrets."
83
+
84
+ if file is None:
85
+ return "Please upload a file to analyze."
86
+ if not query.strip():
87
+ return "Please enter a question or analysis request."
88
+
89
+ # Gradio provides the file object, use .name to get the path to the temporary file
90
+ file_path = file.name
91
+ filename = os.path.basename(file_path) # Extract just the filename from the path
92
+
93
+ file_extension = os.path.splitext(filename)[1].lower()
94
+
95
+ try:
96
+ # 1. Load Document
97
+ documents = get_document_content(file_path, file_extension, filename)
98
+ print(f"Document content loaded. Total {len(documents)} parts/pages/rows identified.")
99
+
100
+ # 2. Split Text (even Excel rows can be split further if they are long)
101
+ text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
102
+ texts = text_splitter.split_documents(documents)
103
+ print(f"Text split into {len(texts)} chunks for embedding.")
104
+
105
+ if not texts:
106
+ return "No usable text extracted from the document after splitting. It might be too short or unparsable."
107
+
108
+ # 3. Create Embeddings and Vector Store
109
+ print("Creating embeddings and vector store (FAISS)...")
110
+ vectorstore = FAISS.from_documents(texts, embeddings)
111
+ print("Vector store created successfully.")
112
+
113
+ # 4. Set up RetrievalQA Chain with Custom Prompt for desired output
114
+ template = """
115
+ You are an expert document analyst. Use the following context to answer the user's question.
116
+ Be thorough, precise, and directly address all parts of the question.
117
+ If the question cannot be answered from the provided context, please state that explicitly
118
+ and explain why (e.g., "The document does not contain information about...").
119
+ Do not make up information.
120
+ The context might contain information from an Excel file, where each entry represents a row.
121
+
122
+ Context:
123
+ {context}
124
+
125
+ Question: {question}
126
+
127
+ Your detailed analysis and answer:
128
+ """
129
+ QA_CHAIN_PROMPT = PromptTemplate.from_template(template)
130
+
131
+ qa_chain = RetrievalQA.from_chain_type(
132
+ llm=llm,
133
+ chain_type="stuff", # "stuff" combines all retrieved docs into one prompt
134
+ retriever=vectorstore.as_retriever(),
135
+ chain_type_kwargs={"prompt": QA_CHAIN_PROMPT} # Pass your custom prompt
136
+ )
137
+
138
+ # 5. Get Answer
139
+ print(f"Running RAG query for: '{query}'")
140
+ result = qa_chain.run(query)
141
+ print(f"Analysis complete. Result length: {len(result)} characters.")
142
+
143
+ return result
144
+
145
+ except ValueError as e:
146
+ print(f"Value error during analysis: {e}")
147
+ return f"Analysis Error: {e}"
148
+ except Exception as e:
149
+ print(f"An unexpected error occurred during analysis: {e}")
150
+ # In a production app, log the full traceback here for debugging.
151
+ return "An internal server error occurred during analysis. Please try again or contact support. Check the Space logs for more details."
152
+
153
+
154
+ # --- Gradio Interface Setup ---
155
+ iface = gr.Interface(
156
+ fn=analyze_document_with_rag, # The Python function to wrap
157
+ inputs=[
158
+ gr.File(label="Upload a Document", file_types=[".txt", ".pdf", ".docx", ".xlsx", ".xls"], type="file"),
159
+ gr.Textbox(label="Your Question/Analysis Request", placeholder="e.g., 'Summarize this document.' or 'What are the key findings?'")
160
+ ],
161
+ outputs=gr.Textbox(label="Analysis Output", show_copy_button=True), # Output for the result
162
+ title="Versatile Document Analysis with RAG",
163
+ description=(
164
+ "Upload a document (.txt, .pdf, .docx, .xlsx, or .xls) and ask a question to get an AI-powered analysis. "
165
+ "For Excel files, each row is processed as a context unit."
166
+ ),
167
+ allow_flagging="never", # Disable flagging system for simple demos
168
+ theme=gr.themes.Soft(), # A modern, soft theme
169
+ examples=[
170
+ # You can add example usage here for users
171
+ # For files, you'd need to provide paths to example files
172
+ # ["path/to/example.pdf", "What is the main topic?"],
173
+ # ["path/to/example.xlsx", "What is the average price of product X?"]
174
+ ]
175
+ )
176
+
177
+ # This block is for running the Gradio app locally.
178
+ # On Hugging Face Spaces, the 'app.py' file is automatically run by the Gradio SDK.
179
+ if __name__ == "__main__":
180
+ if not os.getenv("OPENAI_API_KEY"):
181
+ print("\n" + "="*80)
182
+ print("WARNING: OPENAI_API_KEY environment variable not found for local execution.")
183
+ print("RAG functionality will not work locally without it. Please set it in a .env file or your environment.")
184
+ print("Example: OPENAI_API_KEY='sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx'")
185
+ print("="*80 + "\n")
186
+ iface.launch()