LawCoach

A student can miss a legal issue, cite the wrong rule, or know the rule but struggle to apply it to the facts. A single score doesn't explain which of those happened. I built LawCoach around that distinction, so a practice answer leads to specific feedback and another chance to work on the weak spots.

I'm the sole designer and engineer. I've built the study flows and the AI grading system, along with the database, subscriptions, and deployment. The work spans what a student sees after submitting an essay and what happens when an AI request fails halfway through grading.

• The paid essay panel gives five AI reviewers different jobs. They check issue coverage, legal rules, application to the facts, exam structure, and counterarguments. Each reviews the answer independently before a sixth AI pass combines the reports into a score and a priority improvement plan.
• Grading has to cope with incomplete reviews. The system checks structured responses and counts distinct models that returned usable feedback. If the final synthesis fails, it falls back to a merge of the specialist reports. It stops grading when too few reviewers succeed.
• Practice carries information forward. The app records weaknesses from graded answers and can use them to shape later questions. Students can ask follow-up coaching questions and revise an answer, so the feedback feeds into their next attempt.
• The bar simulator follows a two-day exam structure with six essays, two performance tasks, and 200 multiple-choice questions. I built section timers and saved attempt state to track progress through the exam.
• The study app also includes a diagnostic assessment, flashcards, and practice history. Supabase stores each student's work, while Stripe handles subscriptions and the billing portal.

• LawCoach is in private beta.
• I've been building and maintaining it since 2025.

Next.js 16 · TypeScript · Tailwind CSS · Supabase (Postgres + Auth) · Vercel AI SDK via Vercel AI Gateway · Stripe (checkout, portal, webhooks) · Vercel