Bob Leaf
← The Work Boundlia logo

Children's AI company

Boundlia

Formerly Safe AI Talking Toys · Bounded

Helping companies define, test and manage the boundaries of AI in children's products.

Bob’s role
Co-founder
Founder & CEO
Evren Yiğit
Status
Early stage · in development

Why it matters to him

He holds that for children under 13 “the only safe option is a bounded system,” and that good enough is not good enough where a child is concerned.

What it is

Boundlia is an early-stage company building bounded AI for children’s products. Bob co-founded it with Evren Yiğit, who leads it. Evren states the problem: “Companies want to add AI to children’s products, but need to know whether it will stay within approved content, respond appropriately for the child’s age and handle personal information responsibly, even during long, unpredictable conversations.”

The answer, in the company’s own words: “Boundlia is being developed to help companies define those boundaries, test them and manage them in their products.” The customers are companies: toy makers, schools, publishers and developers. Children and families are the end users.

How it works

Most AI products start from an open-ended assistant and add rules about what it must not say. Boundlia starts by defining the product’s scope: what the AI may draw on, what it may remember, what it may do. It is designed to hold whatever model sits underneath to that scope, with the scope approved by the company and, where memory is concerned, controlled by parents.

Six controls organize the design: what the AI knows, what it remembers, what it does, what it declines, what it records, and how it is evaluated, with testing before launch. The principle is “bounded by design”: the boundary is meant to be part of how the system is built, not a filter laid on top. “Make the boundary structural,” as the company puts it.

What that can and cannot do is worth saying plainly. Some boundaries can be built into the architecture: what data the system can reach, what it can store, what actions it can take. What a language model says can be narrowed and tested, but not guaranteed. That is why testing before launch is part of the design.

Where it stands

The company is early. Boundlia Inc. was formed in Delaware in October 2026, and boundlia.com went live on 12 October. A first phase of research was completed over the summer; the second, now under way, is focused on a working proof of concept. The platform itself is still being developed; early demonstrations have been scripted illustrations, not a working product. Independent testing is a future milestone, not a claim.

The name has changed as the idea matured. It began in early 2026 as Safe AI Talking Toys. The research showed that the larger opportunity lay in the underlying bounded platform rather than in toys, so the company took a working name in September 2026 and became Boundlia in October. Evren: “the broader opportunity did not come from moving away from children. It actually came directly from trying to solve the problem for children.”

Bob’s part

Bob is a co-founder. The under-13 conviction is his; the company is Evren’s to run. Early on he wrote: “MERELY MY IDEAS ON PAPER. Evren is the CEO and will finalize everything.”

The conviction is about children under 13: “I strongly believe that for users under age 13 the only safe option is a bounded system.” He wants education, not distraction: “We are better off insisting on education for children, with games reinforcing that education.”

Separately, Bob is a principal funder of Keep AI Safe, an independent nonprofit that tests children’s AI products. He wants any testing or certification of children’s AI to stay independent of companies he is involved in, including this one, and he said so from the first talks about a toy company: “We need the certification and the consumer digest to be completely arms length from the Talking Toy Company.”

This company is at an early stage. Its products are still being developed. Nothing here is a claim about results.