I Built an AI Mock Trial Platform Because Practicing Law Shouldn't Require a Full Cast

If you're a law student preparing for mock trial, or a lawyer rehearsing for court, you face a frustrating reality: you can't practice alone. A real trial involves a judge, opposing counsel, witnesses, and jurors. To run even a basic practice session, you need to coordinate 3-5 people's schedules. Most of the time, that just doesn't happen. So what do people actually do? The fundamental bottleneck isn't skill — it's access to practice partners. I asked a simple question: what if AI could play every other role in the courtroom? Not a chatbot that answers legal questions. Not a document tool. A full courtroom simulation where: At the end, you get scored on your performance. The challenge isn't just "make AI talk like a lawyer." It's:

1. Multi-role coherence The judge, opposing counsel, and witness are all AI — but they need to behave as separate people with different goals. The judge is neutral. Opposing counsel is adversarial. The witness has a backstory and may be unreliable. One model, multiple conflicting personas, in the same conversation.
2. Stage management A trial isn't a free-form chat. It has strict procedural stages. You can't cross-examine during opening statements. The AI needs to enforce courtroom rules while still feeling natural.
3. Reactive complexity If you raise an objection, the judge must rule. If sustained, opposing counsel must rephrase. If you introduce surprise evidence, the witness must react consistently with their backstory. Every action cascades.
4. Solo practice must feel real If it feels like talking to a chatbot, lawyers won't use it. The responses need enough unpredictability and pushback to create genuine practice pressure. Start with the state machine, not the prompts. I spent too long tweaking AI personalities before realizing the real problem was managing trial flow. Once I built a proper stage system (8 stages, with rules for what's allowed in each), the AI behavior fell into place. Scoring creates motivation. Early testers would quit mid-trial. Adding a verdict with performance scoring changed everything — people now complete full trials because they want to see their score. Credits > subscriptions for this audience. Law students are broke. A generous free tier with credits lets them actually use the tool. Power users (practicing attorneys) will pay when they see the value. If you're curious: It's free to sign up and run several full trials without paying. I'm a solo developer building this actively — feedback from anyone (devs, lawyers, or just people curious about legal AI) is genuinely welcome.