Skip to content
Takshan, dark ground A capital T built from workshop plates: an ink plate scribed with a brass construction circle and crosshair, a marigold plate scored with a fold line, a lacquer dovetail keyed across the seam between them, and a chisel drawn in hairline down the stem.
Book a demo

Spec sheet · Takshan-EPT · AI education

The chapter is open. The drafting is done.

A teacher opens the chapter and the platform drafts the lesson's working material from it. A student gets a tutor scoped to the same chapter. The loop closes where it should: with the teacher's signature.

Designation
Takshan-EPT
Class
AI education
Status
Flagship
Drawing
Sheet 03
One action
The working session
A heavy book lying open at blank cream pages under a workshop lamp; a walnut rule across the gutter, a marigold marker and a pair of dividers on the page.
The open chapter is the whole scope — the tutor answers from it, and the teacher signs what stands

Three seats at one chapter

Three seats at one chapter — the assessment loopSheet 03 · scale 1:1
  1. 01The teacher

    Eight kinds of teaching output generated from the open chapter, ready to edit rather than write.

  2. 02The student

    Nine study tools and a tutor whose scope is the open chapter — a boundary built as retrieval against the book, and one you can probe yourself in the session.

  3. 03The institution

    Class analytics that name the chapter to reteach, and the appraisal record for career advancement under UGC norms.

The loop

Bank → paper → marked → signed.

Assessment runs from question bank to set paper to marked script, and nothing reaches a student's record until the teacher signs it.

Takshan-LM

The model stays home.

The tutor does not have to phone anyone’s cloud. Takshan-EPT can run against a language model deployed inside the POD itself — built on Sarvam-M under the Apache 2.0 licence and tuned on curriculum content used only with permission. Student questions, student answers and the model they meet all stay inside the institution’s own walls. It runs today; the working session can put it in front of you.

Languages

Seven languages, as shipped — not two plus a roadmap.

Deployment (spec, not adoption)

One POD — the platform's self-contained deployment unit — serves 200 concurrent users; three PODs scale to 600. The full POD specification, hardware and storage model included, is published on the Takshan-EPT engineering page at synaptron.ai.

Tolerances

No accuracy figure is published for the tutor — its boundary is a mechanism, and the session is where you probe it. Student-data handling and residency are session questions until their published sheet lands. The concurrency figures above are infrastructure sizing, not adoption. The local model’s lineage is stated, not hidden — Sarvam-M, Apache 2.0, curriculum-tuned; no benchmark figure is printed for it, for the same reason as the tutor.

The material before the drafting — eight kinds of teaching output begin here, ready to edit rather than write

See a chapter taught, studied and assessed.

Thirty minutes across all three seats — teacher, student, institution.

Book the working session