Tim DraperFounder and Managing Partne, Draper Associate · I back the ideas that look crazy first

Qazi AI is aiming to revolutionize justice ⚖️

2 Apr 2025 · 2 min

Watch on YouTube

1:36

In this panel discussion, Tim Draper and several speakers explore Qazi AI’s strategy to encourage users to choose its quasi (alternative dispute resolution) platform over traditional courts. They identify three primary adoption drivers: faster resolution without long court queues, perceived impartiality that avoids judge bias, and the ability to evaluate litigation costs in real time. Concerns about bias in large language models are raised, prompting Qazi AI to explain its technical safeguards: the platform runs on Meta’s open‑source LLaMA model, leverages Meta’s built‑in guardrails, and incorporates proprietary “Negron” features designed to reduce bias. The team also highlights its partnership with Pakistani law firms and legal experts to ensure compliance and credibility as it launches in the Pakistani market. Overall, the conversation underscores both the user‑centric benefits and the technical‑legal risk mitigation measures underpinning Qazi AI’s push to revolutionize justice.

Do you think that people will just use quasi instead of going to court?
That is exactly the idea.
So you think people will do that, though? That's a change of attitude. And usually they're mad at each other.
Yes.
What makes them say, okay, we'll go to quasi?
Yeah, it was full of interruptions, like Tim does it.
But this time I think we were more prepared for this.
2 reasons. First of all, you don't have to sit in a long line where you don't get any resolution and your grandchildren are coming for resolution. Your grandchildren are coming.
Exactly, exactly.
I've had similar disputes for land record in my hometown. And impartiality. So you're not worried that this judge is in a bad mood or for any bias, gender, race, is going to be biased against me.
And there is a third one as well. Before going into the litigation, when you are in the heat of the moment, you can evaluate the costs. That do I even stand a chance or not?
You brought up bias. I was going to ask because Bias in LLMs, large language learning models, is a well-documented problem. What are you doing to mitigate errors and bias in your models, and who has the legal expertise that's advising your team about how to navigate this?
So we are starting off with Pakistani market. So we have law firms and some legal experts on board who are advising us on this. Right now, our model is based on Meta's LLaMA, an open-source model. So we are extensively using the guardrails provided by Meta. And some of our own features that we call Negron features to make sure that the model is overall unbiased.