app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModel
import torch
import torch.nn.functional as F
import json
from typing import List, Dict


def load_model_and_tokenizer():
    tokenizer = AutoTokenizer.from_pretrained("models/tokenizer")
    model = AutoModel.from_pretrained("models/model")
    return tokenizer, model


def load_data(file_path: str = "data.json") -> List[Dict]:
    with open(file_path, "r") as f:
        data = json.load(f)
        flattened_courses = []
        for course_category in data["courses"]:
            for subcourse in course_category["subcourses"]:
                if not subcourse or not all(
                    key in subcourse for key in ["name", "description", "link"]
                ):
                    continue

                flattened_courses.append(
                    {
                        "course_type": course_category["course_type"],
                        "name": subcourse["name"],
                        "description": subcourse["description"],
                        "link": subcourse["link"],
                        "content": f"{course_category['course_type']} - {subcourse['name']}: {subcourse['description']}",
                    }
                )
        return flattened_courses


def get_embedding(text: str, tokenizer, model) -> torch.Tensor:
    inputs = tokenizer(
        text, return_tensors="pt", truncation=True, max_length=512, padding=True
    )
    with torch.no_grad():
        outputs = model(**inputs)

    attention_mask = inputs["attention_mask"]
    token_embeddings = outputs.last_hidden_state
    input_mask_expanded = (
        attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    )
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(
        input_mask_expanded.sum(1), min=1e-9
    )


def precompute_embeddings(documents: List[Dict], tokenizer, model) -> torch.Tensor:
    embeddings = []
    for doc in documents:
        embedding = get_embedding(doc["content"], tokenizer, model)
        embeddings.append(embedding)
    return torch.cat(embeddings, dim=0)


def semantic_search(
    query: str,
    doc_embeddings: torch.Tensor,
    documents: List[Dict],
    tokenizer,
    model,
    top_k: int = 3,
) -> List[Dict]:
    query_embedding = get_embedding(query, tokenizer, model)
    similarities = F.cosine_similarity(query_embedding, doc_embeddings)
    top_k_indices = torch.topk(similarities, min(top_k, len(documents))).indices

    results = []
    for idx in top_k_indices:
        doc = documents[idx.item()]
        results.append(
            {
                "course_type": doc["course_type"],
                "name": doc["name"],
                "description": doc["description"],
                "link": doc["link"],
            }
        )

    return results


try:
    print("Loading model and tokenizer...")
    tokenizer, model = load_model_and_tokenizer()

    print("Loading documents...")
    documents = load_data()
    if not documents:
        raise ValueError("No valid courses found in the data file")

    print("Precomputing embeddings...")
    doc_embeddings = precompute_embeddings(documents, tokenizer, model)
    print("Initialization complete!")
except Exception as e:
    print(f"Error during initialization: {str(e)}")
    raise


def search_interface(query: str) -> str:
    if not query.strip():
        return "Please enter a search query."

    results = semantic_search(query, doc_embeddings, documents, tokenizer, model)

    output = "# Search Results:\n\n"
    for i, result in enumerate(results, 1):
        output += f"## {i}. {result['course_type']} - {result['name']}\n"
        output += f"**Description:** {result['description']}\n"
        output += f"_[Link to Course]({result['link']})_\n\n"

    return output


app = gr.Interface(
    fn=search_interface,
    inputs=gr.Textbox(
        lines=2,
        placeholder="Enter your search query here (e.g., 'machine learning', 'python for beginners', 'deep learning')",
    ),
    outputs=gr.Markdown(
        value="## Search Results will be displayed here.",
        line_breaks=True,
        label="Search Results",
        show_label=True,
    ),
    title="Analytics Vidhya Course Search Engine",
    description="Search for courses using semantic similarity. Results are ordered by relevance.",
    allow_flagging="never",
)

app.launch()