A map of the Guru's connections.
ShabadVerse turns Sri Guru Granth Sahib Ji into an interactive graph. Every one of the 5,542 shabads is a star; the threads between them trace shared themes, so you can wander from one shabad to its kin and discover where a thought travels across Gurbani.
How to use it
- Search four ways. Type English transliteration (ABC), first letters from the start of a line (ੳ▸) or from anywhere in it (▸ੳ▸), or just describe what you are thinking of in plain words (✨) and let meaning search find the closest shabads.
- Browse the Constellation Map. Open TAGS to see the full theme vocabulary; filter it, pick a theme, and jump to any shabad that carries it.
- Expand to explore. Tap a shabad to make it the center; its neighbors fan out by shared theme. Follow the trail and your path stays in the breadcrumbs above.
- Build a library. Add shabads as you go, then switch to REVIEW to read them in full. Save and share your collection with a link.
Reading the graph
- The bright center is the shabad you are exploring now.
- Green dots are shabads already in your library.
- Pale dots are neighbors, colored by their primary mood when one stands out.
- Theme Amber words floating in the field are theme labels; click one to see every shabad under it.
- A thicker thread means a stronger thematic match.
How the connections are built
The threads are not hand-drawn and they are not guesses from a single model. Each shabad's themes come from a consensus of four independent language models, and the graph is grounded in the shabads the Panth already sings together.
Qwen3, DeepSeek-R1, Llama 3.1, and Claude Sonnet each read every shabad and propose themes. A theme only survives if at least two of the four agree, which strips out one-model noise.
The canonical tag set is derived first from the shabads in Amrit Keertan, the keertan the Panth holds in common, then extended to the rest of Gurbani; tradition sets the vocabulary, not the model.
Synonyms collapse into one canonical term, so "Haumai," "ego," and "ego dissolution" count as one idea rather than three thin tags. Themes so broad they fit most of Gurbani are kept off the map: a label on four shabads in five cannot tell two apart.
Two shabads connect on a blend of shared themes and meaning (sentence-embedding similarity). Rare themes count for more than common ones, so a link says why these two belong together, not merely that both speak of the Divine.
Gurbani text, transliteration, and translations come from BaniDB, an initiative of the SikhiToTheMax project.
Straight talk about the AI
- What the AI did. It read translations and proposed themes; four models cross-checked each other. It also powers meaning search, which compares your phrase to each shabad as a vector. The app itself was built with Claude Code.
- What it did not do. It did not write, edit, or reinterpret a single line of Gurbani. The scripture is BaniDB's; the AI only labels and links.
- Where it can be wrong. Themes are machine judgments and some will miss the mark. Treat them as a way to navigate, not as a teaching. When a tag feels off, it probably is, and your eye on the original shabad is the final word.
- Tell us. Found a bad connection or a missing theme? Open an issue on GitHub. The taxonomy gets better every time someone does.