opensearch-natural-language-python-cdk-example

Query OpenSearch indices with natural language.

This is an example application that showcases LLMs ability to translate natural language into OpenSearch/ElasticSearch queries that can be used to return results to the user.

The application is written in Python and utilizes AWS via CDK IaC.

1. The user asks a question in natural language.
   • This is hardcoded in the lambda in this example.
2. The lambda calls OpenAI text-davinci model to convert query into an OpenSearch query.
   • OpenAI API key managed in secrets manager rather than environment variables for security.
3. OpenSearch queried using response from text-davinci.
4. User receives OpenSearch documents based on their query.

Examples

OpenSearch movie data:

title	director	year
Moneyball	Bennett Miller	2011
Star Wars: Episode I - The Phantom Menace	George Lucas	1999
28 Days Later	Danny Boyle	2002
Shaun of the Dead	Edgar Wright	2004
The Grand Budapest Hotel	Wes Anderson	2014

Q: Find all movies that were made after 2010

A: Moneyball, 28 Days later, Shaun of the Dead, The Grand Budapest Hotel

Q: Find all movies that were directed by George Lucas with Star Wars in the title

A: Star Wars: Episode I - The Phantom Menace

Caveats / Further Improvements

• Provide more data in prompt - The model has to assume the structure of the documents fields. If it understood the schema then it could be made more generic and extensible.
• Validating input and output - both for the user and the model's output. We wish to avoid giving OpenSearch bad queries and similarly sending user prompts to the language model raw could lead to hijacking.
• Model fine tuning - having a specific language model for open search queries could yield better results.
• Testing, DevOps etc

Development

This project is setup for CDK development with Python.

The cdk.json file tells the CDK Toolkit how to execute your app.

This project is set up like a standard Python project. The initialization process also creates a virtualenv within this project, stored under the .venv directory. To create the virtualenv it assumes that there is a python3 (or python for Windows) executable in your path with access to the venv package. If for any reason the automatic creation of the virtualenv fails, you can create the virtualenv manually.

To manually create a virtualenv on MacOS and Linux:

$ python3 -m venv .venv

After the init process completes and the virtualenv is created, you can use the following step to activate your virtualenv.

$ source .venv/bin/activate

If you are a Windows platform, you would activate the virtualenv like this:

% .venv\Scripts\activate.bat

Once the virtualenv is activated, you can install the required dependencies.

$ pip install -r requirements.txt

At this point you can now synthesize the CloudFormation template for this code.

$ cdk synth

To add additional dependencies, for example other CDK libraries, just add them to your setup.py file and rerun the pip install -r requirements.txt command.

Useful commands

• cdk ls list all stacks in the app
• cdk synth emits the synthesized CloudFormation template
• cdk deploy deploy this stack to your default AWS account/region
• cdk diff compare deployed stack with current state
• cdk docs open CDK documentation
• python -m pytest runs unit tests

Notes

Docker must be installed to bundle the python lambda with the appropriate packages.