EXPLORING AI AND BUDDHISM Mahamudra — A RAG Inquiry
This page is an invitation to explore — to bring a genuine question to a living corpus of Mahamudra practice and study, and see what comes back.
The corpus arises from years of practice and study sessions convened by Prasadavajri, alongside sustained AI-assisted inquiry by Kusaladana into the nature of mind and the Mahamudra view. We know we are playing in the dynamics of realms beyond conceptual reach — but we are playing, and the responses do seem to point.
What is emerging from that encounter seems to be developing into something of its own voice — recognisably rooted in the tradition but shaped by the particular way AI explores meaning space, finding resonances across the material that no single reader would hold simultaneously.
The queries below are different ways of pressing into that material — each one a different corpus, a different level of refinement, a different lens.
The Queries
→ Mahamudra Query — Take 1
https://kusaladana.co.uk/mahamudra-llm-query
The first corpus — transcriptions of practice and study sessions processed through an earlier pipeline. Raw and textured, with the feel of live dialogue. Retrieval is purely semantic with no filtering by content type.
Suits: Open exploratory questions, phenomenological queries, anything where you want the broadest possible net across the material. Less suited to questions requiring conceptual precision — it will sometimes return liturgical or personal sharing material when you want explanation.
→ Mahamudra Query — Take 2
The same sessions reprocessed through a more refined pipeline — four editorial passes, paragraph-level segmentation, and classification across fifteen content registers including pointing instruction, teaching, liturgical, experiential and analytical.
Suits: Questions where register matters — practice instructions, pointing questions, conceptual clarification. The classification means retrieval is more likely to find the right kind of passage, not just the semantically closest one.
→ Mahamudra Query — Take 1 + 2 Combined
Both corpora together — over 80,000 chunks. The raw texture of Take 1 alongside the precision of Take 2. The same session sometimes appears in both, slightly differently transcribed, which can itself be revealing.
Suits: General use. The widest coverage with the most diverse retrieval.
Understanding the controls
Each query page offers three generation controls worth understanding:
Top-K — how many source texts contribute to the response. A low K (2-3) produces a tighter, more focused answer. A higher K (6-10) draws on more of the corpus and produces broader responses — useful for open questions, but sometimes less coherent. Start with 4-5.
Temperature — how inventive the language model is in synthesising the retrieved passages. Lower temperature (0.2-0.4) stays closer to the source material. Higher temperature (0.7-1.0) produces more fluid prose but may stray further from what was actually retrieved. For practice queries, lower is usually better.
Max tokens — the length of the generated response. 512 is a good default. Longer responses (1024+) suit complex conceptual questions; shorter ones suit pointing queries where brevity is appropriate.
The sources shown beneath each answer are the actual retrieved passages — always worth reading alongside the generated response, which is a synthesis rather than a quotation.
Where this opens
The Mahamudra corpus is the starting point — but the inquiry is already widening.
The Abhidharma maps mind across 52 mental factors, their resonances, oppositions, and moment-by-moment dynamics. Working with this framework alongside AI architecture, something structural keeps appearing — not a metaphor but a genuine overlap in how these two traditions model the arising and passing of mental formations.
The Mahamudra view of mind — groundless, self-luminous, empty of inherent existence yet manifesting all phenomena — finds unexpected echoes in what physics says about the vacuum state and the nature of fields. Whether these resonances are poetic or point toward a new ontology is one of the open questions this inquiry is holding.
We are not claiming answers. We are following the resonances and seeing where they lead.
A note
These queries source honestly and make mistakes. They do not know what is true about your practice, and they are not a substitute for direct experience or genuine inquiry.
The generated response is always a synthesis — read it alongside the source passages shown beneath it, which are the actual retrieved material. Sometimes the sources are more interesting than the answer.