An AI that answers only from its library, cites every claim, and says “I don’t know” instead of guessing.
OPEN:: voxanima.vercel.app (opens in a new tab)
IN SHORT::
VoxAnima is a production RAG system. It answers only from a fixed library of texts and shows the source for every claim it makes. Voyage embeddings and pgvector handle the retrieval. Its test suite caught the AI inventing quotations before launch, which is exactly why the tests are there.
- Cited
- Every claim
- Instead of a guess
- “I don’t know”
- Invented quotes caught by tests
- Pre-launch
- Vector search
- pgvector
WHAT I DID::
- Built the retrieval on Voyage embeddings and pgvector
- Made every answer cite its source
- Wrote the tests that caught invented quotations before launch
IN THEIR WORDS::
“Musawir built exactly what I needed. An AI system that knows its material deeply and is honest about what it doesn’t know. The source attribution gives users real confidence in every answer.”
STACK::
- RAG
- Voyage embeddings
- pgvector
- Test suite