How eMMMy learns to ignore side talk in Aryma Dimension through Jev

We at Aryma Labs and Aryma AI have been at the forefront of applying cutting edge AI solutions.

Recently we were asked, "Do we use Jev?"

Yes, we do. And here is how we used it.

One of the interesting challenges we encountered while building eMMMy inside Aryma Dimension was something surprisingly human.

How do you teach an AI agent to ignore side talk?

Imagine you're in a marketing strategy meeting.

Someone says:

"John, the YouTube contribution looks high. eMMMy, can you show me the ROAS of YouTube?"

Now, this sentence contains two different conversations.

One is directed at John, a colleague.

The other is directed at eMMMy, our AI agent.

A conventional voice-enabled AI assistant might struggle to distinguish between the two. It could interrupt conversations, respond to comments not meant for it, or mistake a passing reference for an instruction.

And in a live meeting, that can become annoying very quickly!

This is where Jev comes in.

Jev is the contextual routing layer we've built to help eMMMy distinguish between conversations directed at the agent and conversations happening among humans.

Here's how it works.

Stage 1: Understand the room

Jev doesn't interpret a sentence in isolation.

It considers who is speaking, what was said previously, the people involved, the current slide, and the ongoing conversation.

Stage 2: Determine intent

Jev asks a narrow but important question:

"Is this meant for eMMMy or for the people in the room?"

It evaluates the context and assigns a verdict with confidence.

Stage 3: Decide whether to intervene

There are three possible outcomes:

• For eMMMy: Respond with the requested data, charts, or insights.
• For the people: Stay silent and let the conversation continue.
• Unsure: Don't rush to respond.

Stage 4: Sometimes, waiting is intelligence

When Jev is uncertain, it waits briefly (2 seconds) to see whether another participant responds.

If a colleague answers, eMMMy stays out.

If nobody responds and the context indicates the request was directed at eMMMy, it can proceed.

This seemingly small capability addresses a much larger problem in agentic AI.

Most AI assistants are designed to answer questions. But an AI agent participating in human conversations must also know when to remain silent.

And that distinction matters enormously when AI becomes an active participant in marketing strategy discussions, rather than merely a chatbot waiting for prompts.

At Aryma AI, we're building Aryma Dimension around this philosophy.

Not just AI that understands marketing measurement. But AI that understands the dynamics of a meeting.

Feel free to schedule a demo of Aryma Dimension.