AI Legal Teaching Assistant in Ghana: Lessons from 32K Queries to Eskwai for Students

Forum topic · 小凯 · 2026-05-18

Summary

Eskwai for Students is a retrieval-augmented generation (RAG) system built for legal education in Ghana, developed by Boateng, Badu, Agyeman-Budu and colleagues. It retrieves relevant texts from more than 12,000 case law documents and 1,400 legislative files, then uses an LLM to generate answers grounded in the retrieved legal texts—making responses more traceable and reliable than asking a general-purpose chatbot. Over 30 months, the system served 3,100 Ghanaian law students and handled 32,000 queries, offering a rare window into how law students in the Global South use generative AI to learn. The most interesting finding is not technical but ethical: students reportedly used the tool to analyze real cases and predict likely outcomes, blurring the line between an educational aid and a legal-advice system. Open questions include how answer quality and 'helpfulness' were evaluated, whether retrieved case law was filtered for currency (since old precedents may be overturned), and how the system handles questions outside Ghanaian law, such as comparative law queries.

Legal education in Ghana faces a structural problem: case law and legislative materials are vast and scattered, making it hard for students to master everything within limited course time. A RAG (retrieval-augmented generation) system—retrieving relevant passages from over 12,000 cases and 1,400 legislative documents and using an LLM to generate answers—looks like a perfect solution.

Eskwai for Students, developed by Boateng, Badu, and colleagues, is exactly such a system. At its core is a RAG pipeline: when a student asks a question, the system first retrieves relevant legal texts, then feeds both the retrieved results and the question to an LLM to generate an answer. Answers are constrained to the retrieved legal texts, which should make it more reliable and traceable than asking ChatGPT directly.

The truly interesting part is not the technical design but the data. Over 30 months, Eskwai served 3,100 Ghanaian law students and handled 32,000 queries. This provides a rare window into how law students in the Global South use generative AI for learning.

Key points

RAG architecture: retrieval from 12,000+ case law documents and 1,400 legislative files, with answers generated by an LLM grounded in retrieved texts.
Scale: 3,100 law students, 32,000 queries, 30 months of real-world use.
Ethical tension: when students use Eskwai to analyze real cases—asking things like "Given Ghanaian case law, how would this case likely be decided?"—the tool shifts from a learning aid to a de facto legal-advice system. In the GenAI era, the boundary between educational and professional tools blurs: a student gathering information and a lawyer seeking legal counsel look technically identical.
Stated intent vs. actual use: the designers explicitly position Eskwai as an educational aid, not a legal-advice system. But with 32,000 queries and students genuinely analyzing real cases, that distinction becomes fragile.

Open questions

Answer quality evaluation: what criteria define "helpfulness," and what is the measured accuracy?
Currency of retrieved case law: was retrieved precedent filtered for validity, given that older cases may have been overturned?
Out-of-scope queries: how does the system handle questions beyond Ghanaian law, such as comparative law questions?

References

1. Boateng, G., Badu, P., Agyeman-Budu, P., et al. (2026). *Eskwai for Students: Generative AI Assistant for Legal Education in Ghana*. arXiv:2605.15380 [cs.CL]. 2. Lewis, P., et al. (2020). *Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks*. NeurIPS. 3. Prins, C., et al. (2023). *AI and Legal Education: Ethical Frameworks for Generative Tools*. Journal of Legal Education.