app.py

import json
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
from transformers import pipeline
import gradio as gr

# Load your natural-language corpus
with open("electricity_corpus.json", "r") as f:
    corpus = json.load(f)

# Build TF-IDF index
vectorizer = TfidfVectorizer()
tfidf_matrix = vectorizer.fit_transform(corpus)

# Load the QA model
qa_pipeline = pipeline("question-answering", model="distilbert-base-cased-distilled-squad")

# Function to retrieve top matching rows
def get_top_contexts(question, top_k=3):
    question_vec = vectorizer.transform([question])
    similarities = cosine_similarity(question_vec, tfidf_matrix).flatten()
    top_indices = similarities.argsort()[-top_k:][::-1]
    return [corpus[i] for i in top_indices]

# Main logic to get answer
def answer_question(question, top_k=3):
    if not question.strip():
        return "Please enter a valid question."

    contexts = get_top_contexts(question, top_k)
    combined_context = " ".join(contexts)[:4096]  # truncate to model max input
    result = qa_pipeline(question=question, context=combined_context)
    return result["answer"]

# Gradio interface
iface = gr.Interface(
    fn=answer_question,
    inputs=gr.Textbox(label="Ask your question about electricity usage..."),
    outputs=gr.Textbox(label="Answer"),
    title="🔌 Electricity Data Q&A",
    description="Ask questions like 'What was the price for residential in Texas in Jan 2001?' or 'Which state had highest revenue in Jan 2001?'",
)

# Run the app
if __name__ == "__main__":
    iface.launch()