Want to turn your (elastic) search into something as hot as Bing + ChatGPT? Look no further than the Elasticsearch + GPT3 Answerer! Our program intercepts Elasticsearch results and sends them to GPT3 to provide accurate and relevant answers to your queries. Plus, it's just plain fun to use!
- Intercept Elasticsearch results and send them to GPT3 for more accurate answers
- Two installation options: all-in-one and on-the-fly
- Live demo available to see the program in action
It is designed to help users get more accurate and relevant answers to their queries, by leveraging the power of Elasticsearch and GPT3.
Check out our live demo at https://es-gpt.sung.devstage.ai/ to see the Elasticsearch + GPT3 Answerer in action! Please note that the site may be unstable and we are currently using the text-ada-001 model for proof of concept, so the GPT answer may be poor. However, this demo shows the concept of how the Elasticsearch + GPT3 Answerer works.
See this diagram.
To use the Elasticsearch + GPT3 Answerer, you'll need to have access to both Elasticsearch and GPT3, as well as Python installed on your system. We offer two installation options:
To use the all-in-one installation, follow these steps:
Clone this repository to your local machine.
$ git clone https://github.com/hunkim/es-gpt.git
$ cd es-gptModify the .env for your Elasticsearch and GPT3 credentials and crawl_index.py file to index your documents.
$make crawlThen, this will run the backend server:
$ make runThen, visit the backend server. The web page will then intercept the Elasticsearch results and send them to GPT3 to provide a reasonable answer. This method is very fast, as the program embeds documents during indexing.
To use the on-the-fly installation, follow these steps:
Add the following scripts to your search page. See static/p.html. Specify the query, results, and gpt_answer output div IDs in your original search page:
<script src="sse.js"></script>
<script src="es-gpt.js"></script>
<script>
window.addEventListener("load", function () {
summarizeOnChange("query", "results", "gpt_answer");
});
</script>Modify the .env with your Elasticsearch and GPT3 credentials. Install the required dependencies by running the following command in your terminal:
$ make run
Run your search web enter a query. The program will intercept the HTML results and send them to GPT3 to provide a reasonable answer. This method is convenient, but slower, as the program embeds the search results and query on-the-fly.
We welcome contributions from the community! If you have ideas for how to improve the Elasticsearch + GPT3 Answerer, please open an issue or submit a pull request. We love hearing from fellow search enthusiasts!
This program is licensed under the MIT License