OUR SOLUTIONS

Projects

AI Assistant

Our data query assistant helps attorneys, academic researchers, policymakers, and the general public, easily study prison sentences being served in the California Department of Corrections and Rehabilitation (CDCR). Users can generate cohorts of individuals that meet the conditions they are interested in with a few natural language prompts. These criteria could be related to age, gender, controlling offenses, prior convictions, sentence length, time served, sentencing county, etc.

Details:

We actively work to ensure that our models and tools do not exacerbate existing biases in our society. Our AI assistant is currently in development and will be available to a select group of users for beta testing in March 2025. If you are interested in using our tool, and offering feedback, we welcome you to join our testing user group. Click here to apply.

Our tool

1. Summary

Our AI Assistant is a no-code platform to support public defenders, prosecutors, judges, academic researchers, and community advocates—regardless of their technical background—in exploring our
database of 95,000 prison sentences. By entering simple natural language prompts, users can identify cases of interest based on various criteria such as age, gender, controlling offense, prior convictions, sentence length, time served, sentencing county, etc.

SYNOPSIS:

Process:

User query: “Show me everyone sentenced for robberies.”

Variable and parameter identification: An LLM, called via the OpenAI API, identifies the relevant column (e.g., Offense Description), extracts search terms (e.g., theft, robbery), and determines the logical connector (e.g., OR).

Use Cases

2.1. Find Sentencing Disparities

A dynamic visualization generated by our AI assistant in response to the user query: Find individuals serving time for an offense that can be described as a robbery or theft. The chart shows that there are more Black individuals are over-represented in this category of robbery-related offenses compared to all other ethnic groups.

2.2. Find Similarly Situated Cases

Users can categorize penal codes into various pre-defined types or they can create a new offense type. These classifications allow users to generate scenarios or cohorts of similarly-situated individuals, i.e. those with the same profiles of past and current commitments.

Citation

If you make use of our dataset(s) or code, please cite our work as follows:
APA style:
BibTex:

License

All content in this project is licensed under GNU AGPLv3  License: AGPL v3

You must give appropriate credit to our work. You may not use our work for commercial purposes, which means anything primarily intended for or directed toward commercial advantage or monetary compensation.