src/app.py

import time
from io import BytesIO

import pandas as pd
import streamlit as st
from gliner2 import GLiNER2

# ---------------------------------------------------------------------------
# Constants
# ---------------------------------------------------------------------------

PERSONAL_FIELDS = [
    "Person Name", "Email Address", "Phone Number",
    "Street Address", "City", "Country", "Date of Birth",
]
PROFESSIONAL_FIELDS = [
    "Company Name", "Department", "Job Title",
    "Office Location", "Employee ID", "Skills", "University",
]
BUSINESS_FIELDS = [
    "Counterparty", "Contract Value", "Effective Date", "Jurisdiction",
    "Governing Law", "Invoice Number", "Product Name", "Project Name",
]
ALL_PREDEFINED_FIELDS = PERSONAL_FIELDS + PROFESSIONAL_FIELDS + BUSINESS_FIELDS

MODEL_ID = "fastino/gliner2-base-v1"
EXTRACTION_THRESHOLD = 0.4

# ---------------------------------------------------------------------------
# Page config & styles
# ---------------------------------------------------------------------------

st.set_page_config(
    page_title="AI Excel Entity Extractor",
    page_icon="🔍",
    layout="centered",
)

st.html("""

    <style>

    .stApp { background-color: #fcfcfc; }

    div.stButton > button:first-child {

        width: 100%;

        border-radius: 8px;

        height: 3.5em;

        background-color: #2563eb;

        color: white;

        font-weight: bold;

        border: none;

    }

    div.stButton > button:hover { background-color: #1d4ed8; border: none; }

    .footer { text-align: center; color: #64748b; font-size: 0.85rem; margin-top: 50px; }

    </style>

""")

# ---------------------------------------------------------------------------
# Cached resources & helpers
# ---------------------------------------------------------------------------

@st.cache_resource(show_spinner="Loading AI model…")
def load_model() -> GLiNER2:
    return GLiNER2.from_pretrained(MODEL_ID)


@st.cache_data(show_spinner=False)
def load_excel(file) -> pd.DataFrame:
    return pd.read_excel(file)


def to_excel_bytes(df: pd.DataFrame) -> bytes:
    buf = BytesIO()
    with pd.ExcelWriter(buf, engine="openpyxl") as writer:
        df.to_excel(writer, index=False)
    return buf.getvalue()


def parse_custom_labels(raw: str) -> list[str]:
    return [c.strip() for c in raw.split(",") if c.strip()]


def is_valid_text(value: str) -> bool:
    return bool(value.strip()) and value.lower() != "nan"

# ---------------------------------------------------------------------------
# UI - Header
# ---------------------------------------------------------------------------

st.title("🔍 AI Excel Entity Extractor")
st.markdown(
    "Automatically extract specific entities like Name, Email, etc., "
    "from your spreadsheet text using GLiNER2 Zero-Shot AI."
)

# ---------------------------------------------------------------------------
# Step 1: Upload
# ---------------------------------------------------------------------------

st.write("### 1. Source Data")
uploaded_file = st.file_uploader("Upload an Excel file (.xlsx)", type="xlsx")

if not uploaded_file:
    st.write("### How it works")
    col_a, col_b, col_c = st.columns(3)
    with col_a:
        st.markdown("**1. Upload**\nDrop an Excel file with a column of text (e.g., emails, descriptions, or notes).")
    with col_b:
        st.markdown("**2. Define**\nSelect from common entities like Names and Dates, or type your own custom fields.")
    with col_c:
        st.markdown("**3. Extract**\nThe AI reads every row and creates new columns for every entity it discovers.")
    st.stop()

# ---------------------------------------------------------------------------
# Step 2: Configure
# ---------------------------------------------------------------------------

df = load_excel(uploaded_file)

if df.empty:
    st.error("The uploaded file appears to be empty. Please upload a file with data.")
    st.stop()

row_count = len(df)

st.divider()
st.write("### 2. Configure Extraction")

with st.spinner("Loading configuration…"):
    with st.container(border=True):
        col_select, col_info = st.columns([2, 1])
        with col_select:
            text_column = st.selectbox("Select text column to analyze:", df.columns)
        with col_info:
            st.metric("Total Rows", f"{row_count:,}")

        st.write("---")

        col1, col2 = st.columns(2)
        with col1:
            selected_labels = st.multiselect(
                "Select Fields to Extract:",
                options=ALL_PREDEFINED_FIELDS,
                default=["Person Name", "Company Name"],
                help="Choose common entities from the library.",
            )
        with col2:
            custom_labels_str = st.text_area(
                "Custom Entities (Comma Separated):",
                placeholder="e.g. Case Number, Part ID, Deadline",
                help="Define unique entities specific to your data.",
            )

        active_labels = list(dict.fromkeys(selected_labels + parse_custom_labels(custom_labels_str)))

# ---------------------------------------------------------------------------
# Step 3: Extract
# ---------------------------------------------------------------------------

if not st.button("🚀 Extract Fields"):
    st.stop()

if not active_labels:
    st.warning("⚠️ Please select or define at least one entity to extract.")
    st.stop()

model = load_model()
processed_df = df.copy()
for label in active_labels:
    processed_df[label] = ""

status = st.empty()
progress_bar = st.progress(0)
start_time = time.time()

for i, row in processed_df.iterrows():
    text = str(row[text_column])
    if is_valid_text(text):
        try:
            results = model.extract_entities(text, active_labels, threshold=EXTRACTION_THRESHOLD)
            for label, found_list in results.get("entities", {}).items():
                processed_df.at[i, label] = ", ".join(found_list)
        except Exception as e:
            st.warning(f"Row {i + 1} skipped due to an error: {e}")

    progress_bar.progress((i + 1) / row_count)
    status.text(f"Extracting fields from row {i + 1} of {row_count}…")

duration = round(time.time() - start_time, 1)
progress_bar.empty()
status.empty()

st.success(f"✅ Extraction complete - {row_count:,} rows processed in {duration}s.")

st.write("### 3. Extraction Preview")
st.dataframe(processed_df.head(10), use_container_width=True)

st.download_button(
    label="📥 Download Enriched Excel File",
    data=to_excel_bytes(processed_df),
    file_name="AI_Extracted_Report.xlsx",
    mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
)

# ---------------------------------------------------------------------------
# Footer
# ---------------------------------------------------------------------------

st.markdown("---")
st.markdown(
    '<div class="footer">Powered by '
    '<a href="https://github.com/fastino-ai/GLiNER2" target="_blank">GLiNER2</a>'
    " • Open-source Zero-Shot Named Entity Recognition</div>",
    unsafe_allow_html=True,
)