Artificial intelligence
Cyntia Intelligence

The rules are the product. The model is replaceable.

Intelligence runs cognitive processes: sets of rules that frame any language model and put it to work on a file, a recording or a piece of text. The same process can be operated by hand, called from a service, or invoked by a step inside Workflow — which is how a flow ends up with judgement in it.

What a cognitive process actually does

Something comes in — a file, a recording, a piece of text. It goes through the rules that frame the model, the model answers inside them, and what comes out is checked before anything else uses it. The model changes; the process does not.

Called from Workflow a service A cognitive process A file An audio A text Rules Model Check Output fields, a decision
  • The rules frame the model
  • The model only answers inside them
  • One input, moving

Asking a model is not the same as running a process

Input Model A different answer each time

A model on its own

The same document, asked twice, comes back two different ways. When somebody asks why, there is nothing to point at: no rule that was applied, no version that was in force, no record of what the model was told. It works in a demo, and it cannot be put in front of an auditor.

Input Model Output Rules The same answer, and it can be audited

A cognitive process

The rules live outside the model and they are the same for every case. What went in, which rules applied and what came out is written down — so an answer can be reviewed, and the model underneath can be replaced without rewriting the process.

One agent, whichever channel they arrive by

Virtual agents are the part of Intelligence that talks to people. The agent is the same thing everywhere: the same knowledge base, the same tools, the same rules. What changes is the door they came in through.

Phone system call Web call Web chat WhatsApp Other chats A virtual agent Knowledge base Tools Cognitive processes

Each agent takes a single channel: the one on the phone system and the one on WhatsApp are two agents, on purpose — what is said on a call is not written the way a chat is. What they share is everything behind them.

What Intelligence is made of

Seven modules. Everything the product does is one of these, and they are the same whether one process runs ten times a day or ten thousand.

  • 01

    Process

    Where a cognitive process is actually run: what goes in, and the result that comes out. The same thing a service call does, with a screen in front of it.

  • 02

    Cognitive processes

    The rules that frame the model: what it may read, what it has to answer, what shape the answer takes, and what happens when it does not fit.

  • 03

    Virtual agents

    Agents that hold a conversation on one channel, with a flow of their own. They are designed, tested and watched from the same module.

  • 04

    Knowledge base

    The documents an answer may be grounded in — retrieval, so the model answers with what the business wrote and not with what it read on the internet.

  • 05

    Tools

    What the model is allowed to call: a system of record, a web service, a query. Configured here once, and available to every process and every agent.

  • 06

    Reporting

    What ran, how long it took, what it cost and what came out. A process nobody can measure is a process nobody can defend.

  • 07

    Administration

    Who may operate what, which models are in play, and where each process runs.

Bring us something you are still reading by hand.

A stack of files, a folder of recordings, a mailbox. Tell us what has to come out of it and we will tell you whether a cognitive process can do it — and if it cannot, we will say so.

Talk to us