app.py

import gradio as gr
import pandas as pd
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from sentence_transformers import SentenceTransformer

# ✅ Load the model
model = SentenceTransformer("all-MiniLM-L6-v2")

# ✅ Load posts dataset
posts_df = pd.read_csv("posts_cleaned.csv")
post_texts = posts_df["post_text"].astype(str).tolist()

# ✅ Cache embeddings
post_embeddings = None

def get_embeddings():
    global post_embeddings
    if post_embeddings is None:
        print("Computing embeddings...")
        post_embeddings = model.encode(post_texts, convert_to_tensor=False)
    return post_embeddings

# ✅ Main recommender logic with JSON output
def recommend_json(user_text):
    if not user_text.strip():
        return { "data": [[]] }
    
    embeddings = get_embeddings()
    user_vec = model.encode([user_text])
    sims = cosine_similarity(user_vec, embeddings)[0]
    top_idxs = sims.argsort()[-5:][::-1]
    recommendations = posts_df.iloc[top_idxs]["post_text"].tolist()
    
    return { "data": [recommendations] }

# ✅ Gradio JSON interface
interface = gr.Interface(
    fn=recommend_json,
    inputs=gr.Textbox(label="What are you interested in?"),
    outputs=gr.JSON(label="Structured Recommendations"),
    title="🔍 AI Content Recommender (API Mode)"
)

interface.launch()