app.py

import os
import gradio as gr
import pdfplumber
import numpy as np
import faiss
import zipfile
import re
from transformers import T5Tokenizer, T5ForConditionalGeneration
from sentence_transformers import SentenceTransformer

# -------------------------
# Step 1: Unzip documents.zip if needed
# -------------------------
def unzip_docs():
    if os.path.exists("documents.zip") and not os.path.exists("docs"):
        print("🔓 Unzipping documents.zip...")
        with zipfile.ZipFile("documents.zip", "r") as zip_ref:
            zip_ref.extractall("docs")
        print("✅ Extracted to /docs")

# -------------------------
# Step 2: Load PDF Content
# -------------------------
def load_docs(folder="docs"):
    all_text = ""
    files_found = []
    for root, _, files in os.walk(folder):
        for fname in files:
            if fname.lower().endswith(".pdf"):
                path = os.path.join(root, fname)
                files_found.append(path)
                with pdfplumber.open(path) as pdf:
                    for page in pdf.pages:
                        text = page.extract_text()
                        if text:
                            all_text += text + "\n"

    print(f"📚 Loaded {len(files_found)} PDF(s).")
    print("🧾 Sample Text:\n", all_text[:500])
    return all_text

# -------------------------
# Step 3: Chunk text semantically
# -------------------------
def chunk_text(text, max_words=300):
    raw_paragraphs = re.split(r'\n{2,}|\n\s*\d{1,4}\s*\n', text)  # split by double newlines or page numbers
    chunks = []
    current = ""
    for para in raw_paragraphs:
        if len((current + para).split()) < max_words:
            current += " " + para.strip()
        else:
            chunks.append(current.strip())
            current = para
    if current:
        chunks.append(current.strip())

    clean_chunks = [chunk for chunk in chunks if len(chunk.split()) > 30]
    print(f"📦 Total Chunks: {len(clean_chunks)}")
    for i, chunk in enumerate(clean_chunks[:3]):
        print(f"🔹 Chunk {i+1}:\n{chunk[:300]}\n")
    return clean_chunks

# -------------------------
# Step 4: Build FAISS RAG Index
# -------------------------
def build_index():
    unzip_docs()
    raw_text = load_docs("docs")
    global doc_chunks
    doc_chunks = chunk_text(raw_text)
    embeddings = embedder.encode(doc_chunks, convert_to_numpy=True, normalize_embeddings=True)
    rag_index = faiss.IndexFlatIP(embeddings.shape[1])
    rag_index.add(embeddings)
    return rag_index

# -------------------------
# Step 5: RAG-based Answer Generator
# -------------------------
def generate_answer(question):
    q_embed = embedder.encode([question], normalize_embeddings=True)
    D, I = index.search(np.array(q_embed), top_k)

    retrieved = [f"Passage {i+1}:\n{doc_chunks[i]}" for i in I[0]]
    context = "\n\n".join(retrieved)

    prompt = (
        "You are KatibaGPT, an expert legal assistant trained on the Constitution of Kenya. "
        "Use ONLY the following legal text to answer the question. Your response must be accurate, structured, and clearly reference the Constitution.\n\n"
        f"Context:\n{context}\n\n"
        f"Question: {question}\n\nAnswer:"
    )

    input_ids = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1024).input_ids
    output_ids = model.generate(
        input_ids,
        max_length=700,
        num_beams=6,
        temperature=0.6,
        repetition_penalty=1.2,
        early_stopping=True
    )

    return tokenizer.decode(output_ids[0], skip_special_tokens=True)

# -------------------------
# Step 6: Load Everything Once
# -------------------------
print("⏳ Loading KatibaGPT backend...")
embedder = SentenceTransformer("multi-qa-MiniLM-L6-cos-v1")
tokenizer = T5Tokenizer.from_pretrained("google/flan-t5-large")
model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-large")
top_k = 5
doc_chunks = []
index = build_index()
print("✅ KatibaGPT is ready.")

# -------------------------
# Step 7: Gradio UI
# -------------------------
demo = gr.Interface(
    fn=generate_answer,
    inputs=gr.Textbox(label="Ask KatibaGPT", placeholder="e.g. What does Article 43 say about the right to housing?"),
    outputs=gr.Textbox(label="Answer"),
    title="⚖️ KatibaGPT — Kenyan Constitution Legal Assistant",
    description="Ask questions about the Constitution of Kenya. KatibaGPT retrieves specific clauses and answers clearly using grounded legal text.",
    examples=[
        "What is the process of impeaching a president in Kenya?",
        "Explain the right to education under the Constitution.",
        "Which article talks about the role of the judiciary?"
    ]
)

demo.launch()