Skip to content
How it works

Intelligence thinks. Intuition decides.

Psychologists describe two ways of thinking.1,2 System 1 is fast and automatic: you recognize a face, read a road sign, know a link looks wrong. System 2 is slow and careful: you check a contract, work out a sum, weigh a hard case. Most decisions are System 1. System 2 steps in when something needs thought.

Artificial Intuition

A maze becomes a reflex.

A maze is a chain of small decisions: left or right, at every turn. System 1 takes each turn as it comes. System 2 stops and works each one out. Both get through. For the thousands of everyday decisions your systems make, the reflex is what you want, as long as it says when a turn is hard.

System 1

VAKYN

takes each turn at a glance

System 2

an LLM

stops to think at every turn

An animation of two runners solving the same mazes. System 1, VAKYN, takes each turn at a glance and solves maze after maze. System 2, an LLM, stops to think at every turn and solves a few in the same time. The same mazes, the same time. An illustration, not a measurement: every turn is a small decision.
System 1 and System 2

VAKYN is System 1 for your systems.

It decides the everyday cases in milliseconds and says how sure it is. The unsure ones go to System 2, an LLM or a person.

Expert intuition becomes reliable through practice and feedback in a familiar setting.3 The VAKYN loop is that practice and feedback, for your work: we find where VAKYN gets your cases wrong, train exactly there, and check again, on demand.

Intelligence thinks. Intuition decides.

System 1

VAKYN

  • fast
  • automatic
  • everyday decisions
  • says how sure it is

Most cases

Which team should get this ticket?

billing · 97% sureyour code routes it

System 2

an LLM or a person

  • slow
  • careful
  • complex decisions
  • thinks it through

The unsure ones

Does this CV show Kubernetes experience?

unclear · 58% surea person checks

The VAKYN loop · practice and feedback

  1. 01System 2 decides the unsure cases
  2. 02we find where VAKYN gets your cases wrong
  3. 03train exactly there
  4. 04check again, on demand

The VAKYN loop is that practice and feedback, for your work.

Two columns. System 1 is VAKYN: fast, automatic, everyday decisions, and it says how sure it is. System 2 is an LLM or a person: slow, careful, complex decisions, it thinks them through. The cases VAKYN is unsure about go from System 1 to System 2. System 2's decisions come back into VAKYN as the VAKYN loop: we find where VAKYN gets your cases wrong, train exactly there, and check again, on demand. Example values are illustrative.
The flow

One request. Three steps.

You send a state and typed questions about it. VAKYN sends back one answer per question, with how sure it is. Your code turns that into a decision.

request.json
{  "model": "jev-latest",  "state": "I was charged twice for my March invoice."}
Step I of III

The same flow on a real document, with real answers: the worked example.

Question types

How you ask it

You give VAKYN a state and questions with fixed answers: yes or no, pick one, give a score. In the API these are noul, choice and score. Start from what you want to know. The type decides the shape of the answer. Ask as many questions as you need in one request, of any type.

What are you asking VAKYN?

  • noul

    Yes or no?

    Use for
    A fact or a rule that is either true or false about the state.
    Example
    “Is this product review fake?”
    Returns
    The probability of yes, one number from 0 to 1.
    More about noul

  • choice

    Which one?

    Use for
    Picking one of a few named options: an action, a queue, a category.
    Example
    “Approve, flag or reject this expense report?”
    Returns
    The pick, a probability for each option and a confidence.
    More about choice

  • score

    How much? What level?

    Use for
    Placing the state on an ordered scale: risk, urgency, severity.
    Example
    “How risky is this expense report?”
    Returns
    The expected level, a probability for each level and a confidence.
    More about score
Pick the question type from what you are asking: noul for yes or no, choice for which one, score for how much or what level.
Under the hood

Read once, answer every question.

The state is read once. Each question branches from that reading. A small head scores the options, and each answer is calibrated, so ten questions about one document cost little more than one. No text is generated, so there is nothing to parse.

State

Read once

The model reads the state one time and keeps that reading in a cache.

cache

  • fake noulHead: scores yes and noCalibrated: for noul, 2 optionsyes 0.04
  • topic choiceHead: scores 5 topicsCalibrated: for choice, 5 optionsquality · 0.97
  • negative scoreHead: scores 5 levelsCalibrated: for score, 5 levels3.94 of 4

Each question continues from the cached state, so only its own words are read.

Runs onVAKYN MAX, hosted by usOr your own GPU or CPU, with open UQFF weights (Apache-2.0)Self-hosted, nothing leaves the machine
Inside one request: the state is read once and cached; each question branches from the cache; a small pointer head scores each option from the model's hidden states; each answer is calibrated for its question type and number of options; it runs on VAKYN MAX or, with the open weights, on your own hardware.

See it on real requests

Notes

  1. 1System 1 and System 2 were named by Keith Stanovich and Richard West: Individual differences in reasoning, Behavioral and Brain Sciences 23(5), 2000. ↩
  2. 2And made famous by Daniel Kahneman, Thinking, Fast and Slow, Farrar, Straus and Giroux, 2011. The two systems are a way to describe two modes of thinking, not two parts of the brain. ↩
  3. 3When intuition can be trusted: Daniel Kahneman and Gary Klein, Conditions for intuitive expertise: a failure to disagree, American Psychologist 64(6), 2009. ↩