Spaces:
Running
Running
title: LLMSearchEngine | |
emoji: 🏆 | |
colorFrom: gray | |
colorTo: purple | |
sdk: docker | |
app_file: app.py | |
pinned: false | |
# LLM Search Engine | |
This is a Flask-based web application that uses a large language model (LLM) to generate search engine-like results, styled to resemble Google’s classic search results page. Instead of querying an external search API, it prompts an LLM to create titles, snippets, and URLs for a given query, delivering a paginated, familiar interface. | |
## Why We Built It | |
We created this app to explore how LLMs can mimic traditional search engines by generating results directly from their training data. It offers: | |
- A nostalgic, Google-like pagination design with clickable links. | |
- A proof-of-concept for LLM-driven search without real-time web access. | |
- A simple, self-contained alternative for queries within the model’s knowledge base. | |
## Features | |
- **Google-Styled Interface**: Search bar, result list, and pagination styled with Google’s colors and layout. | |
- **Generated Results**: Titles, snippets, and URLs are fully produced by the LLM. | |
- **Pagination**: Displays 10 results per page, up to 30 total results across 3 pages. | |
## Limitations | |
- **Static Knowledge**: Results are limited to the LLM’s training cutoff (e.g., pre-2025). | |
- **Generated Content**: URLs and snippets may not correspond to real web pages—use as a starting point. | |
- **No Real-Time Data**: Best for historical or established topics, not breaking news. | |
## Using It on Hugging Face Spaces | |
### Try the Demo | |
Deployed on Hugging Face Spaces, you can test it at [https://codelion-llmsearchengine.hf.space](https://codelion-llmsearchengine.hf.space): | |
1. Open the URL in your browser. | |
2. Type a query (e.g. "best Python libraries") in the search bar and press Enter or click "LLM Search". | |
3. Browse the paginated results, styled like Google, using "Previous" and "Next" links. | |
## Using It as an API | |
### API Endpoint | |
Your app doubles as an API when hosted on HF Spaces: | |
- **URL:** `https://codelion-llmsearchengine.hf.space/` | |
- **Method:** `GET` | |
- **Parameters:** | |
- `query`: The search query (e.g., `"best Python libraries"`). | |
- `page`: Page number (`1-3`, defaults to `1`). | |
### Example Request | |
```bash | |
curl "https://codelion-llmsearchengine.hf.space/?query=best+Python+libraries&page=1" | |
``` | |
### Response | |
Returns **raw HTML** styled like a Google search results page. | |
--- | |
## Integration | |
You can fetch results programmatically and render or parse the HTML: | |
```python | |
import requests | |
from urllib.parse import quote | |
query = "best Python libraries" | |
page = 1 | |
url = f"https://codelion-llmsearchengine.hf.space/?query={quote(query)}&page={page}" | |
response = requests.get(url) | |
html_content = response.text # Render or process as needed | |
print(html_content) | |
``` | |
--- | |
## How It Works | |
1. **LLM Prompting** | |
- Queries trigger a prompt to the `"gemini-2.0-flash-lite"` model. | |
- Generates **30 results** in JSON format. | |
2. **Rendering** | |
- Flask converts results into a **Google-styled** HTML page. | |
- Includes a **search bar, results, and pagination**. | |
3. **Deployment** | |
- Runs via **Flask** and **Docker** on HF Spaces. | |
- Serves **dynamic pages** based on URL parameters. | |
--- | |
## Setup Locally | |
### Install dependencies | |
```bash | |
pip install -r requirements.txt | |
``` | |
### Set environment variables | |
```bash | |
export OPENAI_API_KEY="your-key" | |
export OPENAI_BASE_URL="your-url" | |
``` | |
### Run the app | |
```bash | |
python app.py | |
``` | |
Visit `http://localhost:5000`. |