A Sevak, Not a Guru: rebuilding the chain of transmission with AI

Before India wrote its knowledge down, it spoke it. The Vedas, the Upanishads, the great commentaries — for the better part of three thousand years these were carried not in books but in the chain that Sanskrit calls guru-shishya parampara: teacher to student, mouth to ear, one human being at a time. Knowledge passed guru-mukh, from the mouth of the teacher, and the student's task was to receive it without distortion before passing it on in turn.

It is easy to mistake this for a primitive arrangement that writing later improved upon. It was not. The oral transmission of the Vedas, with its elaborate systems of recitation built specifically to detect and correct error, preserved tens of thousands of verses across millennia with a fidelity that many written manuscript traditions never matched. The parampara was, in the most precise sense, a technology — a method for moving a large and exact body of knowledge through time without losing it. It worked astonishingly well.

It had one weakness. It needed a living guru.

The chain is thinning

That weakness is now the problem. The people who hold living, working command of these texts — who can read the Sanskrit, place a verse in its tradition, and explain what it actually means rather than what a translation gestures at — are few, and they are ageing. Each one who passes without a student of equal depth is a link removed from a chain that took three millennia to forge.

Meanwhile the corpus itself has become unreachable to the people who inherited it. And here it is worth being clear about what the corpus is, because the modern instinct is to file all of this under "spirituality" and move on. The classical Sanskrit canon is a civilisation's entire operating record: the Caraka and Sushruta Samhitas on medicine and surgery; Aryabhata's mathematics and astronomy, written in verse, with their own notation for very large numbers; the Arthashastra on statecraft and economics; the Natyashastra on drama and aesthetics; the Yoga Sutras on the architecture of the mind; the Dharmashastras on law. This is not a devotional shelf. It is medicine, mathematics, governance, and art — and it sits behind a language that perhaps a fraction of a percent of its inheritors can read.

So you have a body of knowledge of extraordinary range, a transmission system that is failing for want of teachers, and a billion-odd people for whom it is, in practice, a closed library. That is the gap I built MyGurukul to address.

Why most AI gets this exactly wrong

Large language models can read Sanskrit and discuss it fluently, and that fluency is precisely the trap. The obvious product — a chatbot that answers questions about the Gita — is the opposite of a parampara, and dangerously so.

A parampara prized one thing above all: faithful transmission. Fluency was never the point; accuracy was. A model that confidently produces a plausible-sounding "verse" it has half-remembered, or smooths a difficult teaching into a comfortable platitude, is not a digital guru. It is the failure mode the entire oral tradition was engineered to prevent, now reintroduced and dressed up as wisdom. Invented scripture is worse than silence, because it carries the authority of the real thing while being a fabrication. If AI is going to touch this material at all, the first design decision — before features, before interface — has to be a vow of restraint.

The discipline of the sevak

So MyGurukul is built not as a guru but as a sevak — a servant. The distinction is deliberate and it runs all the way down into the system's instructions. The sevak never generates scripture from memory. It may quote only passages it has actually retrieved from the authenticated corpus, and when it quotes, it shows the Sanskrit original, its IAST transliteration, and the translation together, with the source named. It is forbidden from speaking beyond its texts. When a question strays into territory where the texts offer a perspective but not a prescription — an Ayurvedic question that is really a medical one, say — it says so, and points the seeker to a human.

What it does, it does in the posture of service rather than pronouncement: a seeker brings a question, the sevak searches the corpus, and it arranges what it finds — "As the Gita teaches: karmaṇy evādhikāras te mā phaleṣu kadācana — you have a right to your action, never to its fruits (Bhagavad Gita 2.47)" — into an answer that stays anchored to the source the whole way through. It fetches and it arranges. It does not decree.

This is a humbler thing than the marketing instinct wants. It is also the only honest one. The value is not that the machine sounds wise; it is that the seeker is delivered, reliably and with citation, to words that have survived three thousand years of careful keeping.

What it actually is

In practice there are three ways in. There is a daily Sacred Reading — a verse, its translation, and an interpretation that bridges the teaching to an ordinary modern life, ending in a question to sit with rather than an answer to file away. There is Ask the Sevak, where a seeker's question is answered from the texts, with sources shown. And there is the library itself: more than seventy authenticated works across ten categories, from the Vedas and Upanishads through Ayurveda, the philosophical darshanas, mathematics, and law, browsable and searchable down to the chapter.

Beneath the consumer surface there is a slower, more scholarly project I think of as the real work — a durable concept layer in which the meaning of a term like dharma or ātman is not a fixed dictionary entry but a node that accumulates attestations as more texts are ingested, with human scholars validating each sense rather than letting the machine decide. The aim there is not a better chatbot. It is something closer to a living, verifiable map of classical Sanskrit thought — the kind of reference the tradition's own commentators spent lifetimes building by hand.

What a sevak cannot do

I want to be exact about the limits, because overclaiming here would be its own small betrayal. A sevak is not a guru. It can put the right verse in front of you, in the original, honestly sourced, and that is genuinely useful — but the parampara never claimed that the words alone were the teaching. What the guru transmitted, and what the texts insist cannot be had any other way, is anubhava — direct, lived realisation, conferred in a relationship over time. No retrieval system confers that. Nor does this touch the other half of the old gurukul, where character was formed as deliberately as memory, through years of shared life. That does not digitise, and I would distrust anyone who said it did.

MyGurukul is therefore best understood as a bridge — to the texts, and back toward the human teachers who can still do the part no machine can. It widens the door. It does not replace the room.

That is the whole ambition, and I think it is the right size. The oldest knowledge technology India built was a method for transmitting a great inheritance faithfully across generations. The newest one we have can be made to serve that same end — provided it is built, from the first line, to serve it, and not to speak over it.

Explore MyGurukul at mygurukul.org.