app.py

import gradio as gr
import pandas as pd
import pypdf
import docx2txt
from typing import List
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_core.documents import Document
import evaluate
import os

# Optional: Groq API integration
from groq import Groq

# ======================================
# 🔹 Extract text from uploaded file
# ======================================
def extract_text(file):
    if not file:
        return ""

    filename = file.name
    if filename.endswith(".pdf"):
        reader = pypdf.PdfReader(file.name)
        return "\n".join([page.extract_text() for page in reader.pages if page.extract_text()])
    elif filename.endswith(".docx"):
        return docx2txt.process(file.name)
    elif filename.endswith(".csv"):
        df = pd.read_csv(file.name)
        return df.to_string(index=False)
    else:
        return ""

# ======================================
# 🔹 Build FAISS index
# ======================================
def build_faiss(text):
    splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
    docs = [Document(page_content=chunk) for chunk in splitter.split_text(text)]
    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
    db = FAISS.from_documents(docs, embeddings)
    return db

# ======================================
# 🔹 Generate answer using Groq API
# ======================================
def generate_answer(prompt):
    api_key = os.getenv("GROQ_API_KEY")
    if not api_key:
        return "❌ GROQ_API_KEY not set. Add it in Space Settings → Variables and secrets."
    
    client = Groq(api_key=api_key)
    response = client.chat.completions.create(
        model="llama-3.3-70b-versatile",
        messages=[
            {"role": "system", "content": "You are a factual assistant."},
            {"role": "user", "content": prompt}
        ],
        temperature=0.5,
        max_tokens=500
    )
    return response.choices[0].message.content.strip()

# ======================================
# 🔹 RAG pipeline
# ======================================
def answer_question(file, query):
    if not file:
        return "⚠️ Please upload a document first."
    if not query.strip():
        return "⚠️ Please enter a question."

    text = extract_text(file)
    if not text:
        return "⚠️ Could not extract text from the file."

    db = build_faiss(text)
    retriever = db.as_retriever(search_kwargs={"k": 3})
    retrieved_docs = retriever.invoke(query)
    context = "\n".join([doc.page_content for doc in retrieved_docs])

    prompt = f"""
You are a factual assistant. Use ONLY the provided context to answer.
If the context doesn’t contain enough info, say:
"The document does not provide enough information."
Context:
{context}
Question: {query}
Answer:
"""
    answer = generate_answer(prompt)
    return answer

# ======================================
# 🔹 Optional evaluation
# ======================================
def evaluate_answer(answer, reference):
    if not reference.strip():
        return "No reference answer provided."
    
    # BLEU
    bleu = evaluate.load("bleu")
    bleu_score = bleu.compute(predictions=[answer], references=[[reference]])["bleu"]
    
    # ROUGE
    rouge = evaluate.load("rouge")
    rouge_result = rouge.compute(predictions=[answer], references=[reference])
    rouge1_f = rouge_result["rouge1"]
    rouge2_f = rouge_result["rouge2"]
    rougel_f = rouge_result["rougeL"]
    
    # BERTScore
    bertscore = evaluate.load("bertscore")
    bert = bertscore.compute(predictions=[answer], references=[reference], lang="en")
    bert_f1 = sum(bert["f1"]) / len(bert["f1"])

    return f"""
**Evaluation Results:**
- **BLEU:** {bleu_score:.4f}
- **ROUGE-1 F1:** {rouge1_f:.4f}
- **ROUGE-2 F1:** {rouge2_f:.4f}
- **ROUGE-L F1:** {rougel_f:.4f}
- **BERTScore F1:** {bert_f1:.4f}
"""

# ======================================
# 🔹 Gradio Interface
# ======================================
with gr.Blocks(title="RAG Chatbot AI") as demo:
    gr.Markdown("# 📚 RAG Chatbot AI\nUpload a document and ask a question. Powered by Groq + LLaMA3-70B.")
    
    with gr.Row():
        with gr.Column(scale=1):
            file_input = gr.File(label="Upload PDF / DOCX / CSV")
            question_input = gr.Textbox(label="Your Question")
            answer_btn = gr.Button("Get Answer")
        with gr.Column(scale=2):
            answer_output = gr.Textbox(label="Answer", lines=15)
            reference_input = gr.Textbox(label="Reference Answer (optional)")
            eval_btn = gr.Button("Evaluate Answer")
            eval_output = gr.Markdown()
    
    answer_btn.click(fn=answer_question, inputs=[file_input, question_input], outputs=answer_output)
    eval_btn.click(fn=evaluate_answer, inputs=[answer_output, reference_input], outputs=eval_output)

demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)