JusticeText taps AI to transcribe evidence for public defenders

While studying computer science at the University of Chicago, Devshi Mehrotra and Leslie Jones-Dove were inspired to build tech that centered on the needs of communities historically marginalized by law enforcement. They ended up reaching out to local public defenders, who told them that they were being overwhelmed by hours of jail calls, body cams and other forms of recorded evidence.

As per one estimate, the average officer’s body camera will record about 32 files, 7 hours and 20GB of video per month at 720p resolution. Multiply those figures by the hundreds to tens of thousands of officers in a police force, and it quickly adds up.

“On the one hand, body cams and other devices are critical for holding law enforcement accountable and providing the best defense possible,” Mehrotra told me in an email interview. “However, they exacerbate workload challenges for public defenders, who are facing caseloads 3 to 10 times the recommended amount.”

So Mehrotra and Jones-Dove founded JusticeText, one of the startups participating in the Startup Battlefield 200 at TechCrunch Disrupt 2023. JusticeText is designed to automatically transcribe body cam footage, interrogation videos and more for public defenders, enabling attorneys to take time-stamped notes, create video clips and share evidence with their colleagues.

JusticeText offers a feature that provides several-sentence summaries of each piece of uploaded evidence. Meanwhile, its ChatGPT-powered tool, MirandaAI, allows attorneys to ask free-form questions about their discovery (e.g., “Were any implicit promises made to the client?” or “What is the timeline of events the witness described?”).

“Public defenders owe their clients the best possible defense. However, digital discovery review requires an increasingly large share of limited resources,” Mehrotra said. “Saving discovery review time frees up critical resources that can be spent working the case and building relationships with clients. From a technical perspective, it also means less IT resources dedicated to technical problem-solving — things like figuring out how to play videos in unique proprietary formats or creating video clips for an upcoming trial.”

While that’s all fine and dandy in theory, the idea of uploading evidence to a platform like JusticeText might give some public defenders pause. There’s the risk of exposing evidence to potential data breaches, for one thing. Beyond that, attorneys might rightly be concerned that any uploaded data could be used in ways that they and their clients don’t necessarily consent to, like training JusticeText’s AI models.

I also worry about the accuracy of JusticeText’s transcriptions — particularly given that transcription tech doesn’t tend to perform equally well across different accents and languages. The summaries that JusticeText provides could be off-base as well; nuance isn’t exactly AI’s strong suit.

Mehrotra did her best to allay my fears, claiming that JusticeText only generates a summary for transcripts above a certain confidence threshold and provides a way for users to redact confidential info from transcripts. Where it concerns privacy and data storage, JusticeText — which places data in “secure cloud servers” and encrypts it both in transit and at rest — doesn’t use uploaded data for model training unless it has explicit permission to do so, Mehrotra says.

Those steps have been enough to win over clients, apparently.

In the time JusticeText closed its $2.5 million funding round (which had participation from Bloomberg Beta, True Ventures, LinkedIn co-founder Reid Hoffman, and former Stockton, California, mayor Michael Tubbs), it’s expanded its customer base of public defender offices, which now includes the statewide public defenders systems in Massachusetts and Kentucky.

Mehrotra claims that JusticeText now has a relationship with over 100 public defender agencies, nonprofit service providers and private practice criminal defense firms across the U.S. — and $1 million in annual recurring revenue. That’s a significant bump in deal flow since we last spoke with JusticeText (in September 2022), when the startup had between 50 and 60 partnerships.

Mehrotra tells me that the focus for JusticeText in the near term will be introducing a Spanish to English translation capability and support for “multi-language” recordings — for example, recordings with speech in both English and Spanish. (As AI isn’t a perfect translator, either, I’d hope that the team is careful in implementing this.) JusticeText also plans to grow its team of seven employees to around 10 in the new year, with an emphasis on expanding its marketing and communications function.

“The pandemic has affected our end user by creating a backlog in the criminal court system,” Mehrotra said. “Taken together, this creates a burning platform for change — even if key decision-makers may be harder to reach amid these challenges.”