Mythos Atlas
RAG - Mythology - Verse-Grounded Citations
A RAG platform for exploring Indian and Greek mythology, where answers are grounded in and cite the original source verses rather than generic summaries.
Problem
Ask a general model about mythology and it blends traditions, invents detail, and cites nothing. For source material people care about, an answer without provenance is worse than no answer.
Approach
A RAG platform over Indian and Greek source texts, built on Next.js with Supabase and pgvector for retrieval. The grounding prompt is adapted per tradition so citations point at the unit a reader can actually look up, quoting Bhagavad Gita verses (BG 2.47) rather than page numbers.
Trade-off
Scoping to two traditions rather than covering everything. Norse, Egyptian and the rest would have been easy to add and would have made the demo look broader, but citation quality depends on knowing the source structure well enough to cite it correctly.
Architecture
Next.js frontend with a Supabase backend using pgvector for retrieval. A grounding prompt adapts citations to each tradition's source material, for example citing Bhagavad Gita verses (e.g. BG 2.47) instead of page numbers, so every answer traces back to an original text.
Results
Answers that stay inside the tradition being asked about and cite the verse behind each claim, which is the difference between a mythology chatbot and something a reader can check.
Highlights
- Verse-level grounding and citation
- Indian and Greek mythology coverage
- pgvector-backed retrieval on Supabase
Technologies
- Next.js
- Supabase
- pgvector
- RAG
Screenshots
