CogPath · Cognitive pattern analysis
CogPath extracts cognitive biomarkers from everyday conversation and tracks each person against their own baseline — surfacing subtle change weeks or months before periodic clinical assessment would.
Core principle
We don't compare people to population norms or age-matched cohorts. We establish a personal baseline from the first conversations, then measure every later interaction against it. A person who has always spoken simply isn't flagged — but if their speech simplifies further relative to their own history, that's a meaningful signal.
Grounded in research showing personal deviation from one's own patterns is a more sensitive marker of emerging pathology than cross-sectional comparison to external norms.
What it measures
Every marker is validated in the peer-reviewed literature on speech and cognitive decline.
From the structure and content of speech
From patterns of interaction
From semantic analysis
How it works
No tests, no altered behaviour, no extra interactions required.
The person speaks with their companion as they normally would. Topics are theirs — no tests are administered.
The person's speech is separated from the AI's, so the AI's behaviour never contaminates the cognitive signal.
Linguistic and behavioural markers are computed — some deterministically, some through semantic evaluation.
Each marker is compared to the person's own baseline, producing a standardized deviation and a composite Cognitive Index.
A rolling window evaluates the trajectory — a sustained downward trend, not a single bad call, triggers an alert.
Independent module
CogPath is one of the modules Amigo runs over its conversations — but it's independent. Point it at telemedicine calls, existing voice recordings, or any audio channel that produces speaker-separated transcripts.
Native integration with the Amigo companion platform.
Augment remote visits by extracting biomarkers from the call.
Batch-analyze archived or third-party audio sources.
Engineered for real conversation
Speaker separation, semantic-level analysis and trend-based alerting keep the signal robust to how AI conversation actually behaves.
Temporal metrics use only the person's speaking time as the denominator; participation is based on turn counts, not word counts — independent of how much the AI says.
Analysis works at the semantic level — the meaning and structure of speech — and is explicitly designed to tolerate speech-recognition noise.
Mood and engagement are recorded alongside cognition, and alerts are based on sustained trends across many conversations.
Scientific foundation
Speech-derived biomarkers are an established and rapidly maturing field.
AUC for identifying individuals below normative cognitive thresholds from speech (npj Digital Medicine, 2026).
F1-score discriminating cognitive impairment from healthy controls in conversational speech (CognoSpeak, 2025).
Accuracy distinguishing early Alzheimer's from controls using linguistic features (Frontiers in Aging Neuroscience, 2024).
Positioning
CogPath adds continuous longitudinal resolution between visits — it does not diagnose.
A continuous early-warning layer between scheduled clinical evaluations — providing objective longitudinal data, timely referral triggers, and a higher-resolution record of cognitive trajectory.
The system does not diagnose. It identifies people whose cognitive patterns have changed significantly relative to their own baseline, warranting clinical attention. Currently deployed as a V0 module at a marginal cost under $0.001 per conversation.
CogPath is a screening tool and is not intended for diagnostic use.
We're seeking academic and pharma partners for prospective clinical validation, and care organisations ready to deploy.