OUR CORE TECHNOLOGY
Technology no one can follow
Let's be honest. Cameras, infrared LEDs, apps, servers — given time and money, any company can imitate those.
One thing cannot be imitated: the self-evolving engine we designed in from the very first line of code.
OFFICIAL NAME
Live Evolution
Clinically-Gated Self-Evolution Engine
A closed loop in which AI rewrites its own scoring rules — and a change is adopted only after passing an exam it has never seen.
HOW IT EVOLVES
In plain words: the Answer-Key Loop
Exactly how a student improves with an error notebook — except the one studying is the AI.
Make the answer key
Right after studying, validation-study participants record when they truly focused, zoned out, or did something else. An answer key only a human could know — one that never existed before.
Collect wrong answers
The device's judgments are automatically compared against the key. Every mismatch becomes an entry in the error notebook.
AI fixes the rules
An AI coding agent reads the error notebook and directly edits the judgment rules and thresholds — work a human engineer used to do.
A hidden exam
The revised rules are re-tested on a held-out set the AI has never seen. This is where memorizing answers and truly improving part ways.
Only passers adopted
No metric may regress. Every change — what and why — is permanently logged.
The loop runs automatically with every new answer key. Even while we sleep, the device's accuracy quietly improves.
Why no one can catch up
Our moat is not an algorithm.
It is the factory that keeps beating the algorithm.
① Clinical answer keys
Real labels collected under hospital IRB approval. Copy the hardware and the loop still won't turn without the answer keys — building them means starting from IRB approval all over again.
② Error-notebook know-how
Months of accumulated failure cases and error taxonomies. Knowledge that can't be replicated even if documents leak.
③ Compounding head start
The loop turns with every new data point. The accuracy lead of whoever started first compounds over time — latecomers forever compete with our yesterday.
WHEN AN IMPROVEMENT IS ADOPTED
One adopted improvement arrives in three places at once
The moment a human clicks approve, the new criteria propagate through the whole system — no device recall, no app reinstall.
Server — even the past
Past sessions are retroactively re-scored under the new criteria. This week's improvement makes last week's reports more accurate too.
App — no update needed
The app never computes scores itself — it only reads the server's results. So scoring can evolve without an app update.
Device — from the next session
The in-session real-time response (the return chime's thresholds) is refreshed at the next session start, within safety bounds.
Criteria for irreversible actions like real-time cues are double-clamped to safe ranges on both server and device — evolve fast, intervene carefully.
DESIGNED FOR AI
All the hidden work is done by AI
Re-scoring, error analysis, drafting improvements, re-examination — the invisible repetitive work was designed for AI from day one. Humans do exactly one thing: approve changes that passed the exam.
What the AI does
The moment an answer key arrives: full re-scoring, error-pattern analysis, rule revision drafts, and the hidden-exam re-test — all without human intervention.
What humans do
Review the passing change's evidence and scorecard, click approve. Once a week, check the direction. That's all.
This design is why a small team can run a 50-participant validation study and algorithm improvement at the same time.
HONEST ENGINEERING
We don't inflate the word "AI"
Calling technology by its precise name is where trust begins.