WakiliBench

Inspiration

Legal education often focuses on memorization, while real legal practice demands critical thinking and structured reasoning. Many students lack access to personalized feedback because lecturers cannot review every practice answer in depth. We wanted to build an AI tutor that challenges students the way a law professor would—through questions, feedback, and continuous assessment.

What it does

WakiliBench is an AI-powered legal training platform that helps law students and aspiring advocates improve legal reasoning through deliberate practice.

Users answer realistic legal scenarios, after which AI:

evaluates their response using structured rubrics,
asks Socratic follow-up questions,
explains weaknesses in reasoning,
tracks progress over time.

The platform is designed for education—not legal advice.

How we built it

The application is built with:

Next.js and TypeScript for the frontend
Supabase for authentication and data storage
PostgreSQL for structured assessment data
Qwen Cloud LLMs for legal reasoning, Socratic questioning, and rubric-based evaluation
Tailwind CSS for the UI

The system separates prompt orchestration, evaluation, and progress tracking into independent services, making it easy to support multiple AI providers.

Challenges we ran into

The biggest challenge was ensuring AI behaves like an educator rather than a legal advice chatbot. We redesigned prompts and evaluation workflows so the model focuses on assessing reasoning, identifying missing legal principles, and encouraging deeper analysis instead of generating answers.

Another challenge was balancing response quality with API costs. We built the platform to support interchangeable LLM providers so we can optimize for both quality and affordability.

Accomplishments that we're proud of

Built an end-to-end AI legal assessment workflow
Implemented rubric-based evaluation instead of simple text generation
Created an AI Socratic tutor that encourages critical thinking
Designed a scalable architecture supporting multiple LLM providers

What we learned

Building educational AI requires much more than connecting an LLM API. Careful prompt engineering, structured evaluation, and transparent feedback are essential to create a trustworthy learning experience.