# pdf_qa_app.py import os import gradio as gr from pathlib import Path from langchain_community.document_loaders import PyPDFLoader from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain_community.vectorstores import Chroma from langchain_community.embeddings import HuggingFaceEmbeddings from langchain_core.prompts import ChatPromptTemplate from langchain_core.output_parsers import StrOutputParser from langchain_ollama import ChatOllama # You can swap with ChatOpenAI, ChatGroq, etc. # Set up vector store directory persist_dir = "chroma_pdf" llm = ChatOllama(model="llama3.1:8b", temperature=0.1) parser = StrOutputParser() PROMPT = ChatPromptTemplate.from_template( """You are an assistant that answers using ONLY the context provided. If the answer is not present, say "I don't know." Context: {ctx} Question: {q} Answer with citations like [#] referencing the context. """ ) def process_pdfs(uploaded_files): docs = [] for f in uploaded_files: temp_path = Path("uploads") / f.name temp_path.parent.mkdir(exist_ok=True) with open(temp_path, "wb") as out: out.write(f.read()) loader = PyPDFLoader(str(temp_path)) docs.extend(loader.load()) splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=150) chunks = splitter.split_documents(docs) embedder = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2") db = Chroma.from_documents(chunks, embedding=embedder, persist_directory=persist_dir) db.persist() return "Index built successfully!", True def ask_question(q): embedder = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2") db = Chroma(persist_directory=persist_dir, embedding_function=embedder) docs = db.similarity_search(q, k=4) ctx = "\n\n".join(f"[{i+1}] {d.page_content.strip()} (p{d.metadata.get('page', 0)+1})" for i, d in enumerate(docs)) prompt = PROMPT.invoke({"q": q, "ctx": ctx}) return (llm | parser).invoke(prompt) def reset_index(): if Path(persist_dir).exists(): for file in Path(persist_dir).glob("**/*"): file.unlink() Path(persist_dir).rmdir() return "Index reset. Please upload and rebuild.", False def guarded_ask_question(q, ready): if not ready: return "Please upload PDFs and click 'Build Index' first." return ask_question(q) with gr.Blocks() as demo: gr.Markdown("""# 📄🧠 PDF Q&A Agent Upload PDF(s), build an index, and ask questions with grounded, cited answers. """) uploaded_files = gr.File(file_types=[".pdf"], file_count="multiple", label="Upload PDFs") with gr.Row(): build_btn = gr.Button("Build Index") reset_btn = gr.Button("Reset Index") status = gr.Textbox(label="Status") q = gr.Textbox(label="Ask a question") submit = gr.Button("Submit") answer = gr.Textbox(label="Answer") index_ready = gr.State(False) build_btn.click(fn=process_pdfs, inputs=[uploaded_files], outputs=[status, index_ready]) reset_btn.click(fn=reset_index, outputs=[status, index_ready]) submit.click(fn=guarded_ask_question, inputs=[q, index_ready], outputs=answer) demo.launch()