Getting started

There are two simple ways to get started with the application.

1. Using our UI built using streamlit

2. Python CLI

To install deepsearchai, run

pip install deepsearchai

Run the UI using Streamlit

1. By default, deepsearchai is not shipped with dependencies to render the UI elements. If you wish to use the UI, run

pip install deepsearchai[ui]

2. Create an app instance and run it

from deepsearchai.app import App
app = App()
app.run()

Fig: Add Data to the App via UI

Fig: Query the App via UI

Fig: Response generated

Running the App from the CLI

Initialising the App

Open terminal, create appropriate virtual environment and install the package

pip install deepsearchai

By default, the package is shipped with dependencies to process images. If you also want to process audios, run

pip install deepsearchai[audio]

If you wish to process videos, run

pip install deepsearchai[video]

Open Python Shell and run the following commands

from deepsearchai.app import App

app = App()

Add Data

Local Files

app.add_data(<LOCAL_PATH>)

S3 data from the CLI

app.add_data(<S3_PATH>)

Youtube Channel

Deepsearch lets users index youtube channels. This is a great way to index videos from a channel, and query them later. The indexing happens in a way such that queries will surface the exact second from where the relevant context as per the user query is present

app.add_data("youtube:<YOUTUBE_CHANNEL_ID>")

Deepsearch automatically infers the datasource from the path provided. If the path is a local path, it will be treated as a local file. If the path is a youtube channel, it will be treated as a youtube channel. If the path is an S3 path, it will be treated as an S3 path.

Querying

from deepsearchai.enums import MEDIA_TYPE
app.query(<input string query>, [List of media_types to search across])

For example

from deepsearchai.enums import MEDIA_TYPE
app.query("A car in front of a building", [MEDIA_TYPE.IMAGE, MEDIA_TYPE.AUDIO])

Users can add 3 types of data to the deepsearch

1. Images : This can be images in any standard format (jpg, png, jpeg, etc). Deepsearch will infer the type, and process it accordingly.

2. Audios : This can be audios in any standard format (mp3, wav, etc). Deepsearch will infer the type, and process it accordingly.

3. Videos: This has to be a youtube video channel. Deepsearch has validations which will prevent users from attempting to inject videos in any other format.