Most AI astrology apps invent authority. MyPanditji constrains the model, at the system-prompt level, to quote only from the chunks retrieved for your specific question — and then post-checks every citation before showing you the reply. Invented citations are stripped, not displayed.
The reasoner has no access to text it did not retrieve. Each question pulls 8-12 chunks from the corpus; the reply is constrained to that set.
After the model writes the reply, every (book, chapter, verse) citation is round-tripped against the retrieved chunks. Any citation that does not match an actual chunk is removed before the reply is shown.
Every citation chip in the reply opens the underlying passage with a Sanskrit + English gloss. Verify any claim in two taps.
Question
Why is my Mars in the 10th house considered strong?
Reply
Mars in the 10th carries digbala — directional strength — because the 10th is its directional home. For Karka and Vrishchika lagnas this becomes a yogakaraka, conferring authority but demanding sustained, disciplined action. Both BPHS and Saravali list this placement as karaka-aligned for military, surgical, and engineering vocations.
Tap any citation chip. It opens the source passage from our 87-text corpus. The original Sanskrit + an English gloss appear side-by-side. Verify in two taps.
It cannot. The reply pipeline strips any (book, chapter, verse) reference that does not round-trip to a chunk in the retrieval set for that turn. The user never sees an invented citation.
Yes. Normal, Deep Research, Max Council and Check Your Pandit are all citation-bound. Deep Research and Max produce a bibliography at the foot of the brief in addition to inline chips.
Citation-grounded AI is well-known in academic and search products. Applying it rigorously to a Vedic astrology corpus, with post-checking against retrieved chunks, is novel — most astrology AI products do not do this.
