Practised

How Koher works. What it protects. What it refuses.

The one commitment, turned into working rules. Why the commitment exists is its own page; this is how it is kept, day to day.

why Koher exists →

Judgement Is Personal

When Koher says "humans make decisions," it does not mean humans as a species. It means humans as persons — individuals who have examined their own positions and arrived at them through their own thinking.

Deferring to social convention without examining it is not human judgement. It is cognitive offloading — functionally no different from asking AI to judge for you. The offloading destination changes (culture, tradition, peer consensus, an algorithm), but the abdication is the same. Someone else — or something else — decided, and you inherited the decision without inhabiting it.

This is why Koher’s architecture makes the rules explicit and configurable. Anybody who sits down to write a rule has to work out what they actually think. A discipline supplies one answer and an institution supplies another; the file asks for the person’s own. A teacher setting out how she reads a student’s work is the clearest case.

Most of the people Koher builds for are nowhere near that position. A registrar, a compliance officer, an accounts clerk, a radiographer — the rules arrived before they did and will not be theirs to write. Years of doing the work to somebody else’s specification teach a person that what they notice outside the rule is not knowledge. Koher cannot hand them the file and does not pretend it can. What it offers is smaller: the reasoning behind any verdict it produces stays visible.

Koher does not build tools that judge. It builds tools that make judgement visible — so the person using the tool must decide whether to accept it, revise it, or reject it entirely. That decision, personally made, is what Koher means by human.

The Architecture Is the Proof

The commitment — keep a real outside in the room, and never let an AI take its place — is set out in full on why Koher exists. What matters here is that it is demonstrated in working tools.

Koher builds with a three-layer architecture — qualification, rules, language — where each layer does only what it is suited for. The rules are code and the code is open, so the rules behind any result can be read. Several of the tools also show the working in the product itself.

The AI tool market is full of claims. Tools claim to be transparent, ethical, human-centred. Koher builds tools where transparency is structural: the judgement is code, so it can be inspected, and the rules are explicit, so they can be argued with. Rewriting them is open to whoever sets the procedure where they work, which is not everybody.

Judgement Is Encodable, Shareable, and Plural

A teacher in Ahmedabad can write a JSON config that encodes how she evaluates design coherence. A teacher in Berlin can write a different config. The same student's work, run through both, produces different readings. Neither is wrong. Both are portraits of how those teachers see.

A way of seeing can be written down and passed between people while staying open to disagreement. That is what matters here. The teacher stays where she is; her way of reading becomes portable, and it becomes arguable.

For nearly everybody in that other position, much less of this is available. Somebody whose procedure is not theirs to change will not be writing the file. What reaches them is the result along with the rules that produced it, so the result can be argued with. That is far less than authorship, and it is still the difference between a procedure whose workings are visible and one that issues a number about a person and keeps its reasoning to itself.

Ground Over Surface

Koher has spent enough time around impressive surfaces — institutions with beautiful reputations and hollow classrooms, tools with extraordinary interfaces and no judgement underneath. Koher values work that arrives at the ground level: specific, concrete, honest about what it does and does not know.

A half-finished thing with a genuine question attached is worth more than a polished deliverable that performs completeness. Hesitation is signal. The absence of hesitation means either perfect understanding (rare) or disengagement from difficulty (common). Koher would rather not assume which one it is.

Honesty as Structure

Koher declared its use of AI publicly because it was true. The architecture is transparent because hidden judgement is unauditable judgement. The balance sheet is public because hidden costs become exit ramps.

Koher does not ask anybody to perform enthusiasm. "This tool did nothing for me" is useful and gets read. Nobody is chased for silence.

Where a tool has a Behind the Curtain toggle, you see the result first and can then inspect how it was produced: which layer did what, what signals were read, what rules fired, what the AI narrated. Every tool is open source, so even where there is no toggle the rules behind a result can be read. The commitment is that no part of the judgement is sealed.

Doing More With Less

The prevailing logic of AI development is accumulation: larger models, more parameters, bigger datasets, more compute. Koher inverts this. Smaller models doing remarkable things on constrained hardware — edge deployments, phones, cheap servers, places where a 700 MB model will never run.

If a 50 MB model does what a much larger one does, that is the better design. The constraint is generative: it forces architectural cleverness over brute-force scaling. It means a tool runs on a cheap phone with intermittent connectivity as well as on fibre broadband.

Serve People, Not Institutions

Everything here is free to everybody. Always. No accounts and no email — an invisible bot-check confirms you are a person, and Koher never builds a profile of you. No freemium tiers, no institutional licensing. Somebody working alone in rural Maharashtra and somebody inside a well-funded institution see the same thing, and get the same depth.

When an institution benefits, it is because the people inside it benefit. Koher does not sell to institutions. It builds for people and lets institutions notice.

The Range

Koher began in design education — a teacher building tools for his students. But the architecture illuminates wherever genuine stake exists. Design education, yes. But also: ethical technology, animal rights, artistic practice, philosophy of technology, veganism, critical pedagogy. The range extends wherever the question "what should AI judge, and what should it not?" intersects with a domain where Prayas has lived experience, professional knowledge, ethical commitment, or intellectual investment.

This is not mission creep. It is the natural consequence of an architecture that is domain-agnostic. The three layers work wherever language carries judgement. The specificity is in the configs and the trained models, not in the architecture itself.

Earned, Not Applied For

The people closest to the practice arrive through demonstrated work, not credentials or applications. Contributors are paid per accepted contribution. There is no flat rate for presence. Core team positions — time-based, sustained, with a visible ladder — are earned by people whose work consistently lands at the ground level. The door is open to anyone, nationwide. What matters is the work.

Weird Over Polished

Koher is a teacher making free AI tools with a ten-year horizon, no business model, and a public declaration of how AI is used in its making. None of this is normal.

Koher would rather work with people who find something genuinely confusing or disagreeable about the practice than with people who find it generically exciting. Confusion is engagement. Generic excitement is noise.

Worth Is Not Hierarchical

The hierarchy exists — Prayas sets direction, reviews work, makes architectural decisions. But it is structural, not ontological. A student's observation that "this tool did nothing for me" shaped the practice as much as a year of development.

The Practice Continues Regardless

Koher is a ten-year practice. This is not a tagline — it is a structural decision about what kind of proof is required. A claim this large cannot be proved in a quarter, a funding cycle, or a product launch. It requires tools that accumulate, a body of work that compounds, and enough time for the world to encounter the work and decide whether it matters.

The practice continues regardless of funding, recognition, or adoption. This is not stoicism — it is the removal of exit ramps. When external validation becomes a condition for continuing, it also becomes a reason for stopping. Koher refuses that bargain.

These values are not aspirational. They describe how Koher already works. If something Koher does contradicts what is written here, the contradiction is the problem, not the document. The work itself is the argument.