10 X UR Science

No one knows your data better than you do.

The goal of the URTech Research Agents is to learn from your own data to help you generate queries, perform exploratory analyses, and develop new questions about your data. All of this is done locally and securely within your data silo through the use of lighweight highly specialized AI Agents. The agentic-based LLM tool comes pre-trained in SQL best practices and basic data analytic techniques. With use, it will learn about your data set, the relationships between data tables, and the meaning behind variables and features. Once up to speed and fine-tuned with its own documentation of your dataset, it will be able to turn your questions into queries, and your data into discoveries.

It starts with the Query Generator. Using a local model that has been pre-trained with SQL best practices, this agentic-based system sits on your machine, just like SQL Server or any other data analysis program, and interacts with the data saved on your network, just like you would. By giving the system basic information about how your data is structured, schema names, for example, the system learns how to write queries for your data based on questions that you ask it. An additional output of this system is a repository of its own successful queries, that you have a hand in selecting, that it stores and refers to for future queries. This will increase efficiency and accuracy for pulling data from the right places and with the correct feature names, etc.

From there, the Analysis Engine will take the outputs of the queries and turn them into actual answers to your questions. Using standard data summarization and visualization techniques, this agent can turn the data tables into useful analytics.

The final step is the Interrogation Driver. This unique part of the agent system takes into account the requested queries and output data analytics and develops further hypotheses about the data. It taps into its stored knowledge of the data structures and underlying meaning of the variables to suggest relationships between unexplored variables. With your assistance, it can then posit these questions to the query generator to pull the relevant data. And then it can tap the data analysis engine to display analytics about its question and subsequent data-driven answer.

These agent systems collaborate under an overseer, which connects with the user, to establish priorities with analytics and discovery. Keeping a human in the loop ensures that the system stays on track.