While pursuing computer science at the University of Chicago, Devshi Mehrotra and Leslie Jones-Dove were driven to create technology that addresses the needs of communities historically marginalized by law enforcement. They connected with local public defenders, who expressed their struggles with the overwhelming influx of hours of jail calls, body cam footage, and other forms of recorded evidence.
An estimate indicates that an average police officer’s body camera generates around 32 files, 7 hours, and about 20GB of video per month at 720p resolution. When multiplied by the hundreds or even thousands of officers on a police force, this quickly becomes unmanageable.
“Body cameras and similar devices are essential for holding law enforcement accountable and ensuring the best possible defense,” Mehrotra shared with me in an email interview. “However, they significantly increase the workload for public defenders, who are already handling caseloads three to ten times beyond the recommended limits.”
In response, Mehrotra and Jones-Dove launched JusticeText, a startup. JusticeText is designed to automatically transcribe body camera footage, interrogation videos, and more for public defenders, allowing attorneys to make time-stamped notes, create video clips, and easily share evidence with colleagues.
One of JusticeText's standout features is its ability to provide concise summaries for each piece of uploaded evidence. Additionally, its ChatGPT-powered tool, MirandaAI, enables attorneys to ask free-form questions about their case materials (for instance, “Were any implicit promises made to the client?” or “What is the timeline of events as described by the witness?”).
“Public defenders are committed to providing their clients with the best defense possible. Yet, reviewing digital discovery requires a significant portion of their limited resources,” Mehrotra noted. “By streamlining the discovery review process, we can free up essential resources for case work and client relationships. From a technical standpoint, this also reduces the need for IT resources to tackle technical issues, such as playing videos in unique proprietary formats or preparing video clips for trial.”
Although these innovations prove beneficial, the idea of uploading sensitive evidence to a platform like JusticeText could raise concerns among public defenders. The potential risk of data breaches is a significant factor, as is the possibility that uploaded materials could be used in unintended ways, including training JusticeText's AI models without proper consent.
Moreover, I have concerns regarding the accuracy of JusticeText’s transcriptions, particularly since transcription technology often struggles with various accents and languages. The summaries generated by JusticeText might also lack the necessary nuance; after all, AI is not known for its subtlety.
Mehrotra reassured me that JusticeText generates summaries only for transcripts that meet a specific confidence level and offers users a way to redact confidential information from transcripts. Concerning privacy and data storage, JusticeText employs “secure cloud servers” and encrypts data both during transmission and when stored. Mehrotra emphasized that uploaded data is not used for model training without explicit permission.
These measures have evidently instilled confidence among clients.
Following a successful $2.5 million funding round—with contributions from Bloomberg Beta, True Ventures, LinkedIn co-founder Reid Hoffman, and former Stockton, California, mayor Michael Tubbs—JusticeText has broadened its clientele to include statewide public defender systems in Massachusetts and Kentucky.
Currently, JusticeText has established partnerships with over 100 public defender agencies, nonprofit organizations, and private criminal defense firms nationwide, generating approximately $1 million in annual recurring revenue. This marks a significant increase from when we last spoke in September 2022, when the startup had partnered with around 50 to 60 organizations.
Mehrotra mentioned that JusticeText's immediate focus will be on launching a Spanish-to-English translation feature and support for “multi-language” recordings, which may contain both English and Spanish dialogue. Given that AI translation is not always precise, I hope the team proceeds with caution in this regard. JusticeText also aims to expand its team from seven to around ten employees in the coming year, particularly in marketing and communications.
“The pandemic has created a backlog in the criminal court system that affects our end users,” Mehrotra said. “These challenges create an urgent need for transformation, even if reaching key decision-makers has become more difficult in this environment.”