app.py

# ============================================
# Civil Engineering RAG (ASTM) - Hugging Face Version
# ============================================
import os
import fitz  # PyMuPDF
import faiss
import numpy as np
import gradio as gr
import tempfile
from typing import List
from groq import Groq
from sentence_transformers import SentenceTransformer

# --------------------------
# 🔑 API Key
# --------------------------
GROQ_API_KEY = os.environ.get("GROQ_API_KEY", "")
if not GROQ_API_KEY:
    raise RuntimeError("❌ Missing GROQ_API_KEY. Please add it in Hugging Face → Settings → Secrets.")

# Initialize Groq client and embedding model
client = Groq(api_key=GROQ_API_KEY)
embedder = SentenceTransformer("all-MiniLM-L6-v2")

INDEX, CORPUS = None, []

# --------------------------
# 📄 Safe PDF Text Extraction
# --------------------------
def extract_text_from_pdf(file_path: str) -> str:
    try:
        text = ""
        with fitz.open(file_path) as doc:
            for page in doc:
                text += page.get_text("text")
        return text
    except Exception as e:
        return f"Error extracting text from {file_path}: {e}"

# --------------------------
# ✂️ Chunking Function
# --------------------------
def chunk_text(text: str, chunk_size: int = 800, overlap: int = 120) -> List[str]:
    chunks = []
    start, n = 0, len(text)
    while start < n:
        end = min(start + chunk_size, n)
        chunk = text[start:end].strip()
        if chunk:
            chunks.append(chunk)
        start = end - overlap
        if start < 0:
            start = 0
    return chunks

# --------------------------
# 🔢 Build FAISS Index
# --------------------------
def build_faiss_index(paths: List[str]):
    texts, vectors = [], []
    for p in paths:
        text = extract_text_from_pdf(p)
        if text.startswith("Error extracting text"):
            raise RuntimeError(text)
        chunks = chunk_text(text)
        if not chunks:
            continue
        embs = embedder.encode(chunks, convert_to_numpy=True, show_progress_bar=False)
        texts.extend(chunks)
        vectors.append(embs.astype("float32"))

    if not texts:
        raise RuntimeError("❌ No valid text extracted from PDFs.")

    vectors = np.vstack(vectors).astype("float32")
    index = faiss.IndexFlatL2(vectors.shape[1])
    index.add(vectors)
    return index, texts

# --------------------------
# 📤 Rebuild Index from Upload
# --------------------------
def rebuild_index_from_upload(files):
    if not files:
        return "⚠️ Please upload at least one PDF."

    paths = []
    for f in files:
        try:
            # Gradio provides a temp file path automatically (f.name)
            if hasattr(f, "name") and os.path.exists(f.name):
                temp_path = f.name
            else:
                # fallback in rare case
                with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp:
                    tmp.write(f.read())
                    temp_path = tmp.name
            paths.append(temp_path)
        except Exception as e:
            return f"❌ Error while saving uploaded file: {e}"

    try:
        global INDEX, CORPUS
        INDEX, CORPUS = build_faiss_index(paths)
        return f"✅ Successfully indexed {len(paths)} PDF(s). You can now ask questions!"
    except Exception as e:
        return f"❌ Error while building index: {e}"

# --------------------------
# 🔍 Retrieve Context
# --------------------------
def retrieve_context(query: str, top_k: int = 4) -> str:
    if INDEX is None:
        return "⚠️ Please upload and index PDFs first."
    q_emb = embedder.encode([query], convert_to_numpy=True).astype("float32")
    distances, indices = INDEX.search(q_emb, top_k)
    selected = [CORPUS[i] for i in indices[0] if 0 <= i < len(CORPUS)]
    return "\n\n---\n\n".join(selected)

# --------------------------
# 🧠 Query via Groq LLM
# --------------------------
SYSTEM_PROMPT = (
    "You are a helpful Civil Engineering assistant. "
    "Use ONLY the provided ASTM or uploaded document context to answer. "
    "If the answer isn't in context, say you cannot find it."
)

def ask_groq(query: str, top_k: int = 4, model: str = "llama-3.3-70b-versatile") -> str:
    if INDEX is None:
        return "⚠️ Please upload PDFs first."

    context = retrieve_context(query, top_k)
    if not context.strip():
        return "⚠️ No relevant information found in the uploaded PDFs."

    prompt = f"""{SYSTEM_PROMPT}

Context:
{context}

Question:
{query}
"""

    try:
        completion = client.chat.completions.create(
            model=model,
            messages=[{"role": "user", "content": prompt}],
            temperature=0.2,
        )
        return completion.choices[0].message.content
    except Exception as e:
        return f"❌ LLM Error: {e}"

# --------------------------
# 🎨 Gradio UI
# --------------------------
def ui_ask(query: str, top_k: int):
    try:
        return ask_groq(query, top_k=top_k)
    except Exception as e:
        return f"❌ Error: {e}"

with gr.Blocks(title="Civil Engineering RAG (ASTM)") as demo:
    gr.Markdown("## 🏗️ Civil Engineering RAG\nUpload ASTM or civil-engineering PDFs, build an index, and ask questions.")
    
    with gr.Row():
        uploader = gr.File(label="📄 Upload PDFs", file_count="multiple", file_types=[".pdf"])
        status = gr.Textbox(label="Status", interactive=False)
    uploader.upload(rebuild_index_from_upload, uploader, status)

    gr.Markdown("---")
    inp = gr.Textbox(label="Your Question", placeholder="e.g., What is the curing time for concrete as per ASTM?")
    k = gr.Slider(1, 10, value=4, step=1, label="Top-K passages")
    out = gr.Textbox(label="Answer")
    btn = gr.Button("Ask")
    btn.click(ui_ask, inputs=[inp, k], outputs=[out])

if __name__ == "__main__":
    demo.launch()