Deploying PDF Version v14

app.py

@@ -1,8 +1,7 @@
 import os
 import tempfile
 import logging
-from typing import List
-
+from typing import List, Dict
 from langchain.embeddings import OpenAIEmbeddings
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain.vectorstores import Chroma
@@ -27,20 +26,22 @@ os.environ["CHAINLIT_CONFIG_DIR"] = tempfile.gettempdir()
 APP_CONFIG = {
     "model_name": "gpt-4o",
     "temperature": 0,
-    "chunk_size": 4000,
+    "chunk_size": 2000,  # Increased for book-length content
     "chunk_overlap": 200,
-    "max_file_size_mb": 50,
+    "max_file_size_mb": 100,  # Increased for larger books
     "max_output_tokens": 4096,
     "embedding_model": "text-embedding-3-large"
 }
 
-# Initialize text splitter
+# Initialize text splitter with more sophisticated settings
 text_splitter = RecursiveCharacterTextSplitter(
-    chunk_size=APP_CONFIG["chunk_size"], 
-    chunk_overlap=APP_CONFIG["chunk_overlap"]
+    chunk_size=APP_CONFIG["chunk_size"],
+    chunk_overlap=APP_CONFIG["chunk_overlap"],
+    length_function=len,
+    separators=["\n\n", "\n", " ", ""]  # Prioritize splitting on paragraph breaks
 )
 
-# System template
+# System template (unchanged)
 system_template = """You are GPT-4o, an advanced AI model with multimodal capabilities and high intelligence. Use the following pieces of context to answer the user's question.
 If you don't know the answer, just say that you don't know, don't try to make up an answer.
 ALWAYS return a "SOURCES" part in your answer.
@@ -64,24 +65,37 @@ messages = [
 ]
 prompt = ChatPromptTemplate.from_messages(messages)
 
-def process_file(file: AskFileResponse) -> List[str]:
-    """Process uploaded PDF file and return list of text chunks."""
+def process_file(file: AskFileResponse) -> List[Dict[str, str]]:
+    """Process uploaded PDF file and return list of text chunks with metadata."""
     with tempfile.NamedTemporaryFile(mode="wb", delete=False, suffix=".pdf") as temp_file:
         temp_file.write(file.content)
         temp_file_path = temp_file.name
 
     loader = PyPDFLoader(temp_file_path)
     pages = loader.load_and_split()
-    texts = text_splitter.split_documents(pages)
-    return [text.page_content for text in texts]
+    
+    chunks_with_metadata = []
+    for i, page in enumerate(pages):
+        chunks = text_splitter.split_text(page.page_content)
+        for j, chunk in enumerate(chunks):
+            chunks_with_metadata.append({
+                "content": chunk,
+                "metadata": {
+                    "page": i + 1,
+                    "chunk": j + 1,
+                    "source": f"page_{i+1}_chunk_{j+1}"
+                }
+            })
+    
+    return chunks_with_metadata
 
 @cl.on_chat_start
 async def start():
     """Initialize chat session and process uploaded PDF."""
-    await cl.Message(f"Welcome! Chat session started using {APP_CONFIG['model_name']} for text generation and {APP_CONFIG['embedding_model']} for embeddings. This combination provides state-of-the-art performance for both English and non-English tasks. You can now ask questions about the uploaded PDF.").send()
+    await cl.Message(f"Welcome! Chat session started using {APP_CONFIG['model_name']} for text generation and {APP_CONFIG['embedding_model']} for embeddings. This combination is optimized for processing and querying book-length content. You can now upload a PDF of a book to begin!").send()
     
     files = await cl.AskFileMessage(
-        content="Please upload a PDF file to begin!",
+        content="Please upload a PDF file of a book to begin!",
         accept=["application/pdf"],
         max_size_mb=APP_CONFIG["max_file_size_mb"],
         timeout=180,
@@ -96,11 +110,13 @@ async def start():
     await msg.send()
 
     try:
-        texts = process_file(file)
-        metadatas = [{"source": f"{i}-pl"} for i in range(len(texts))]
-
+        chunks_with_metadata = process_file(file)
+        
         embeddings = OpenAIEmbeddings(model=APP_CONFIG["embedding_model"])
-        docsearch = await cl.make_async(Chroma.from_texts)(texts, embeddings, metadatas=metadatas)
+        docsearch = await cl.make_async(Chroma.from_documents)(
+            [Document(page_content=chunk["content"], metadata=chunk["metadata"]) for chunk in chunks_with_metadata],
+            embeddings
+        )
 
         memory = ConversationBufferMemory(
             memory_key="chat_history",
@@ -116,14 +132,14 @@ async def start():
                 streaming=True
             ),
             chain_type="stuff",
-            retriever=docsearch.as_retriever(),
+            retriever=docsearch.as_retriever(search_kwargs={"k": 3}),  # Retrieve top 3 chunks
             memory=memory,
             return_source_documents=True,
             combine_docs_chain_kwargs={"prompt": prompt}
         )
 
         cl.user_session.set("chain", chain)
-        msg.content = f"Processing `{file.name}` done. You can now ask questions!"
+        msg.content = f"Processing `{file.name}` done. You can now ask questions about the book!"
         await msg.update()
 
     except Exception as e:
@@ -143,8 +159,8 @@ async def main(message: cl.Message):
         source_documents = res["source_documents"]
         
         text_elements = []
-        for source_idx, source_doc in enumerate(source_documents):
-            source_name = f"source_{source_idx}"
+        for source_doc in source_documents:
+            source_name = f"Page {source_doc.metadata['page']}, Chunk {source_doc.metadata['chunk']}"
             text_elements.append(cl.Text(content=source_doc.page_content, name=source_name))
         
         source_names = [text_el.name for text_el in text_elements]