BARBRI GPT · AI study companion

SQE1 is a multiple-choice exam of brutal breadth. Hundreds of topics across two papers, thousands of practice questions, and the same painful moment every time: you get one wrong, read a static explanation, and still don’t understand why your reasoning failed.

A new wave of AI-first prep tools in the UK proved students want something better than static question banks. BARBRI GPT is our answer: an AI study companion built around BARBRI’s question bank, where the tutor sits inside the practice session instead of in a separate chat tab. I designed it and built the front end.

You can’t just bolt a chatbot onto exam prep. Hallucinated law is worse than no help at all, and students preparing for the hardest exam of their lives don’t have spare trust to give. The real question was where AI belongs in a high-stakes study loop, and where it has to be fenced out.

That gave us hard rules. The AI must be grounded in the exact question on screen. Its answers must be labelled as AI-written and easy to check against the official answer. Students need a one-tap way to report a bad explanation. And someone human has to see those reports. Trust is a workflow, not a disclaimer.

I designed this one in code. Working in Vue 3 and TypeScript meant every screen was stress-tested by real state from day one: what happens when a session is refreshed halfway through, what the dashboard looks like with zero data, where focus goes when a modal closes. Empty states, loading states and error states were designed as first-class screens, not afterthoughts.

The AI layer went through the same discipline. Explanation formats, suggested prompts and the plain-English mode were all tested with authored legal content behind the same interface as the live model, so the tutor’s behaviour was designed and reviewed by people who know the law before a single generated answer reached a student.

The tutor sits beside the question, not over it. The session view holds three things in one glance: the syllabus tree, the SRA-style question, and BARBRI GPT. The panel opens with grounded prompts like “explain the tricky parts” and “what legislation applies?”, and it always carries the same notice: written with AI, check it against the official answer.

BARBRI GPT changes how students relate to wrong answers. Instead of reading a static explanation and moving on, they interrogate the question until it makes sense, then pin what they learned as a flashcard. In testing, that loop is where the product earns its keep: the moment of failure becomes the moment of learning.

The trust architecture does its job too. The AI is usable becauseit’s checkable, and the report queue gives the legal team a live view of where explanations need work, which feeds directly back into content quality.