Arsen
Dzantiev
← All projects

02Own productIn development

Thales

Connect with humanity's wisdom.

Thales is for curious people without a philosophy degree. You talk to fifteen philosophers, take any claim apart in five layers and argue with other people in a debate that runs by rules. Marx, the first persona built in full, answers from 19 of his own works and shows where he looked.

The product is in development, about 60 per cent built, and its AI sits behind a login. The public part is a landing film in six scenes and a waitlist, so the output below is shown as examples, written in the product's own format.

My part

End to end: product, design, web development, the AI, art direction and the film pipeline.

Visit the site: thales.institute↗
The first scene of the landing film, live at thales.institute.Live product

In numbers

15
philosophers in the Chamber
19
of Marx's works behind his answers
5
layers in every claim analysis
6
scenes in the landing film

What I built

01

Claim Archaeology

Any claim, taken apart in five layers: what holds it up, where it came from, where it's true, how truth turns into error, and what survives.

The problem

People accept or reject a claim whole. “Possession wins titles” is true somewhere, sometimes.

How it works

  • Logic: explicit and hidden premises, logical form, assumptions, confidence from 0 to 1.
  • History: who arrived at the idea, when and in what context.
  • Boundaries: where it holds, where it fails, edge cases.
  • Transformation: the original truth, its general form, what got lost on the way, and where it flips.
  • Synthesis: what stays true and what to drop. Five chained calls with a strict schema; later layers build on what the earlier ones found.

Status

  • Built: finished, not in daily use yetThe five calls are built. The results page still shows a sample, so the example here is written by hand.

Also fits

Testing analytics claims and scouting conclusions, reviewing strategy assumptions, checking quotes in media, teaching critical thinking.

Claim Archaeology

“Possession wins titles.”

Source: Football punditry · Context: Arguments about how a team should play

1 of 5 layers expanded|
  1. Premises Extracted

    • P1The team that keeps the ball gives the other side fewer chances.
    • P2Fewer chances conceded and more created add up to more wins over a season.Implicit premise
    • P3Possession causes the advantage rather than following from it.Implicit premise

    Logical Structure

    P1 ∧ P2 ∧ P3 → C

    Hidden Assumptions Uncovered

    • !The ball is kept where it hurts the opponent.
    • !The players are good enough to keep it under pressure.

    Analysis Confidence

    82%

    High confidence in argument structure analysis

Written by hand in the product's five-layer format, on a claim clubs argue about.Example

02

Primary Sources and RAG

AI that looks things up in your own library before it answers, weighs how reliable each source is, and shows where it looked.

The problem

Models invent numbers, names and quotes with confidence. For a research newsroom, a club archive or a company knowledge base, that is not acceptable.

How it works

  • Documents are cut by the kind of text: an aphorism stays one idea, a treatise is split around 600 tokens, the rest around 400. Each chunk carries topic tags and the original terms.
  • Search runs on meaning and exact words, 0.65 to 0.35, multiplied by how reliable the source is: five tiers from 1.00 to 0.65, the weakest off by default.
  • Up to six sources go into an answer, and each is listed under it with the work and the section.
  • A persona has a horizon. By default Marx speaks from his own years, 1818 to 1883; an extended mode lets him reason about later events.

Status

  • Built: finished, not in daily use yetMarx: 19 of his works, up to six sources under each answer.

Also fits

Coaching methodology libraries, medical teams, club archives with a legend who knows only his era, law firms.

The Chamber

Karl Marx

1818-1883

Historical

Why do small clubs keep selling their best players to the rich ones?

Football as a market in players came after my time, so I will not pretend to know its details. I can tell you how I would look at it.

As I wrote in Capital, one capitalist always kills many. Money gathers in fewer hands, and the club without it lives by selling the one thing of value it owns: the future work of its best young men.

Notice what is sold. Not the man, but his capacity to work for a number of years. Ask who owns that capacity and who profits from it, and you will understand this market better than any table of fees.

Sources (2)

  • Source|Capital, Volume IChapter 32
  • Source|The Communist ManifestoChapter I
Written by hand in the Chamber's format: Marx in the default historical mode, his sources under the answer.Example

03

Design

Brand, design system, data on screen, reports and sites. Variants side by side on real content; the winner becomes the standard.

The problem

An interface made of template cards and default fonts looks like every dashboard. When every screen is solved differently, people relearn the product on every page.

How it works

  • Every scene is cut from the same stone: pictures from image and video models matched in colour and brightness by measurement, not by eye.
  • Words are set on the page, never baked into the pictures.
  • Search wide, choose narrow: 30 and then 17 variants of the engraved headline, down to one; the Φ mark picked from two grids of nine.
  • Designed and built end to end by me, with AI tools.

Status

  • Live: running in productionThe landing film, live at thales.institute.

Also fits

Brand systems for clubs, agencies and media. Data-heavy dashboards. Reports that look the same on screen and in PDF. Cinematic launch sites.

The Chamber. Every scene is cut from the same stone, matched in colour to the next by measurement.Live product
The Debate. Words are set on the page, never baked into the pictures.Live product

Also inside

Built with

  • Next.js
  • TypeScript
  • Supabase
  • PostgreSQLpgvector
  • Vercel AI SDK
  • Anthropic
  • OpenAI