Case study, 2026
Constitution.ng: a legal AI that cites the section, in five languages
Constitution.ng puts the Nigerian Constitution in plain words, in five languages. Every answer cites the exact section it comes from, with the verbatim text a tap away, so it informs instead of guessing. I designed and built all of it, from the retrieval pipeline to the interface.
- Role
- Founder and CTO, sole engineer
- Where
- Emelar Republik
- When
- 2026
- Stack
- FastAPI, React, Retrieval-augmented generation, LLM retrieval
The problem with asking a chatbot about the law
A general-purpose language model will answer a constitutional question confidently whether or not it’s right. For legal text that’s the wrong failure: a plausible answer with no source is worse than no answer, because the reader can’t check it.
Retrieval first, citation always
Constitution.ng is built on retrieval-augmented generation. Before the model writes anything, the system retrieves the sections of the Constitution that bear on the question, and the answer is grounded in those sections. Each answer names the section it relied on, and the original wording sits one tap away, so a reader can always check the plain-language version against the law itself.
It answers in five languages, because a constitution that most citizens can’t read in their own language isn’t doing its job.
Stack
A FastAPI service runs the retrieval pipeline and the model calls. The interface is React.