Skip to content

Mandana AI

An AI that argues both sides before it answers

Named for Mandana Mishra, a tradition of dialectical debate in which opposing positions are held until a take can survive both.

Explore Mandana AI
Mandana AI

The method

Five parts. Then you ask your own.

Every answer is built the same way. Not a transcript. A contract: the thesis, the other side, what remains, how sure it is, and one question.

  1. 01Lean

    The thesis-informed answer

    The clearest position that can be defended, stated directly. Not a hedge. Not a lecture.

  2. 02Tension

    The strongest opposing truth

    The cost, the counterpoint, or the other side that still has a claim on reality. Skipped only when there is no real tradeoff.

  3. 03Take

    The synthesis that keeps both

    What remains true from both sides, or how the tension resolves, or why it should stay live. Never a false middle.

  4. 04Confidence

    How much of this is grounded

    Grounded, Inference, Speculative, or Unknown — written as a word in the answer. Certainty is not performed where it is not available.

  5. 05Ask

    One catalytic question

    Exactly one follow-up: the question that most increases understanding. Not a pile of suggestions.

Sign up to ask your own

Why Mandana

A method of holding both sides.

Mandana Mishra (c. 8th century) is remembered in the Advaita tradition as a master of opposing schools of thought, not as a rhetorician who won by force of style. The point of a debate, in that lineage, is to find what remains true when both sides have been stated at full strength.

Mandana the product is built on that constraint. It does not flatten a dilemma into a single confident paragraph. It returns a Lean, a Tension, a Take, a confidence line, and one question.

The full story
Mandana Mishra seated under a tree among manuscripts

Why not a flat answer

Same question. Two epistemics.

Should journals require authors to publish null results?

A flat AI answer

Yes, journals should require authors to publish null results because it reduces publication bias and makes science more reliable.

Mandana

Lean. Yes. A literature that only prints hits is not a map of what was tried. It is a map of what survived the editorial filter.

Tension. Journals are scarce attention. Forcing nulls into the same channel can bury the few results that should change practice, and authors will still under-report sloppy nulls.

Take. Require a registered null to be findable, a short, citable record, without pretending every non-finding deserves the same pages as a decisive result.

Two people reading in a library, one seated at a lamp-lit desk and one standing among the shelves

Use

Intellectual work, not generic productivity.

Two researchers at a lab bench, working together at a laptop among glassware and a microscope

Research & argument-building

Steel both sides of a claim before you commit it to a paper, a brief, or a meeting.

Decision-making under uncertainty

When the data is incomplete and the cost of being glib is high.

Debate & essay prep

Find the hidden question, the best objection, and the line that survives both.

Stress-testing your own opinions

Ask Mandana to argue against you as hard as it argues for you.

Also building

The rest of the house.

Mandana is the flagship. Enterprise, Impact, and Embodied sit beside it — same house, different jobs.

  • Enterprise AI

    Specialized systems for high-value workflows.

    Redactron for document privacy. MarketMind for financial intelligence. Desks where a glib answer is a liability.

    Read more

  • Impact AI

    Public-good tools, starting in the field.

    It starts with शेतमित्र (ShetMitra): Marathi voice and text advice for farmers in Maharashtra.

    Read more

  • Embodied AIComing

    Robots taught from how people already work.

    Record the workers who know the job. Use that video to train robots — the skill stays with the person who has it.

    Read more
Two people on a grassy hill watching a rocket climb past a tree
Try it