app.py

import pandas as pd
import gradio as gr
from transformers import pipeline

MODEL_ID = "mcsqstudio/africa-sector-classifier"

SECTOR_NAMES = {
    "ATX": "Agritech",
    "ETX": "Edtech",
    "HTX": "Healthtech",
    "FTX": "Fintech",
    "REC": "Retail & E-commerce",
    "ERG": "Energy",
    "MFG": "Manufacturing",
    "XSC": "Cross-sector",
}

_pipe = None
_model_error = None


def get_pipe():
    global _pipe, _model_error
    if _pipe is None:
        try:
            _pipe = pipeline("text-classification", model=MODEL_ID, truncation=True)
            _model_error = None
        except Exception as exc:
            _model_error = str(exc)
    return _pipe


def classify(text):
    if not text or not text.strip():
        return pd.DataFrame(), "Please enter a company description."
    pipe = get_pipe()
    if pipe is None:
        return pd.DataFrame(), (
            "Model not ready yet. Run **Sita_Sector_Model_v2_transformers.ipynb** "
            "in Colab to fine-tune and push `mcsqstudio/africa-sector-classifier` "
            "to Hugging Face, then click Classify again.\n\n"
            f"Detail: {_model_error}"
        )
    results = pipe(text.strip()[:2000])
    rows = [
        {
            "Sector": r["label"],
            "Sector name": SECTOR_NAMES.get(r["label"], ""),
            "Confidence": round(r["score"], 4),
        }
        for r in results
    ]
    df = pd.DataFrame(rows)
    top = rows[0]
    headline = (
        f"**{top['Sector']} - {SECTOR_NAMES.get(top['Sector'], '')}** "
        f"({top['Confidence']:.1%} confidence)"
    )
    return df, headline


EXAMPLES = [
    "Mobile payment platform enabling small businesses to accept card payments in Nairobi",
    "Solar-powered microgrids bringing affordable electricity to rural communities in Uganda",
    "Online marketplace connecting farmers directly to buyers and aggregating harvest data",
    "AI tutoring app that personalizes math lessons for secondary school students in Nigeria",
    "Telemedicine service offering remote consultations with licensed doctors in Kenya",
]

with gr.Blocks(title="Sita Sector Classifier") as demo:
    gr.Markdown(
        "# Sita Sector Classifier\n"
        "Classifies an African startup description into one of seven Sita Sector industries. "
        "Fine-tuned DistilBERT on the "
        "[Africa Startup Directory](https://huggingface.co/datasets/mcsqstudio/africa-startup-directory).\n\n"
        "**Sectors:** ATX Agritech - ETX Edtech - HTX Healthtech - FTX Fintech - "
        "REC Retail & E-commerce - ERG Energy - MFG Manufacturing"
    )
    txt = gr.Textbox(label="Company description", lines=4)
    btn = gr.Button("Classify", variant="primary")
    headline = gr.Markdown()
    out = gr.Dataframe(
        headers=["Sector", "Sector name", "Confidence"],
        datatype=["str", "str", "number"],
        interactive=False,
    )
    btn.click(classify, inputs=txt, outputs=[out, headline])
    gr.Examples(examples=EXAMPLES, inputs=txt)

demo.launch()