Using Copilot

Conduit focuses on report generation and automating data analyst tasks.

Example: a daily task where Conduit pulls data from Shopify to determine product costs, gathers marketing data from Facebook and accounting systems, and then compiles all this information to calculate the true ROI for products.

Essentially, Conduit automates the work typically done by data analysts. The AI takes user queries and crafts analysis programs that are then executed on our server, making data analysis more accessible.

Data sources for Copilot

Before you can ask the first question or build a report, you must specify the data sources.

• You can specify a direct connection to a connected application (e.g. Facebook)
• Specify a spreadsheet, either by URL or by uploading and downloading a file
• Select a workflow
• You can specify several data sources at once.

Spreadsheets in Copilot

Conduit can handle data from multiple documents and tabs or from a single document. You can supply an unlimited number of data sources in a single query. Example: ask AI to use data from multiple spreadsheets and data from Shopify.

Please note that our UI allows you to clarify the meanings of specific columns and rows for the AI. This significantly enhances the AI's performance.

How does Copilot work with different tables and data sources

Our system automatically generates connections between tables, but you also have the option to define these relationships manually. Upon integrating a data source, AI analyzes a sample of your data, outlines the fields and the links between tables. Our AI often efficiently identifies primary and foreign keys on its own.

However, if the AI misses a connection, you have the option to manually establish it using plain English.

For instance, if you explain, "The Orders table lists the SKU for each product, which corresponds to the product's price in the Prices spreadsheet," the AI will then autonomously connect the tables for any query related to the product's price. This means users won't have to repeatedly instruct the system to associate specific data points.

Copilot for data science

We employ LLMs to generate Python programs. Theoretically, you can craft a prompt directing the AI to apply specific algorithms using Pandas.

However, crafting such prompts requires the user to have data science knowledge, and it might be more efficient to directly use a Jupyter notebook. Conduit was designed with business professionals in mind.

Another option is to integrate your models with Conduit through Workflows. Simply host your model and provide an HTTP API for access.