HAVEN, by Justice Apex Inc.
Every child is trying to say something.
HAVEN is a swarm-intelligence system built to learn what minimally speaking and nonverbal children are trying to say: their attempted words, their sounds, their mumbling, their movements, and what each one means. It's designed to grow with every child it learns from, toward every child who communicates this way.
HAVEN is built, and its intelligence is running today on recordings regularly fed to it from its first family's child. Its confirmed-word learning and its swarm learning loop are running now. Matching a child's attempts against the shows they watch is built and being brought live, and the movement and mimicry comparison is in development. The next step is a trial with additional families, which requires independent ethics review (an IRB) before any other family's data is collected. This funding round goes toward further research, the equipment and requirements for data gathering, and growing the team with engineers and researchers.
How the intelligence works
1. Confirmed words.
HAVEN finds the sounds a child repeats in ordinary home video, groups them, and learns from each one a caregiver confirms. Every confirmed meaning becomes training data, and the system learns from every confirmation it's given. With trial families, this is how the confirmed database is built.
2. Where the words come from.
Many children learn their first words from the shows and videos they watch. HAVEN holds the transcripts of the programs a child watches and is built to cross-reference the sounds they make when they're trying to say a word or a sentence against what was said on screen, to find the phrase they may be reaching for.
3. Movement and mimicry, in development.
A skeleton model reads body movement and mouth shape from video. We're building HAVEN to compare a child's movements with the adults' in the shows they watch, to see what they're imitating and whether it lines up with what they're trying to say. Mouth shape narrows the possible words: it rules out candidates their mouth couldn't have been forming.
4. Swarm intelligence.
These sources don't vote alone. HAVEN's learning loop records each source's guess before the answer is known, checks it against what a person who knows the child confirms, and recalibrates how much to trust each source from those results. That's how it's built to get better at reading a child over time, and it's designed to share what it learns across families without sharing any child's recordings.
What HAVEN isn't
HAVEN supports families. It does not diagnose, and it doesn't replace a clinician, therapist or teacher.
Patent pending. Four provisional applications were filed June 25, 2026.
Talk with Chad
Start with a quick email introduction. Write to Chad at chad@justiceapex.ai.
Please don't include details about a child or any health information in your email.