Our technology

Purpose-built AI for
cognitive assessment.

We trained a specialized model on 220 neuropsychology papers and 11 normative databases to do one job: interpret the results from our 26-test battery. The reports it generates are scored deterministically, grounded in published research, and designed to be shared with clinicians.

220
Research papers indexed
468
Semantic chunks indexed
7
Normative databases integrated
General AI vs. specialized AI

Why we built our own engine instead of using ChatGPT

General-purpose AI models have no access to normative tables, no concept of what "typical" means for a 34-year-old vs. a 60-year-old, and no ability to identify cross-domain cognitive patterns. They sound confident. They also hallucinate citations. We needed something better, so we built it.

Feature ChatGPT / General AI Neuropsych.io Report
Normative comparison Guesses what "normal" means Compares to published norms by age and gender (N=200-2,200 per test)
Standard scores Can't compute - no normative tables Standard Score (M=100, SD=15), percentile, classification band
Cross-domain pattern analysis ~ Describes each score separately Identifies patterns across domains (e.g., low numerical cognition + intact reasoning = dyscalculia profile)
Dyscalculia screening No validated framework MMR framework - separates dyscalculia from math developmental delay
Research citations Hallucinated or generic references Real papers from 220-paper corpus, cited per section
Consistency Different answer each time Deterministic scoring + structured generation = reproducible reports
Validity indicators Takes all scores at face value Flags potentially invalid results (distraction, misunderstanding)
Everyday examples ~ Generic descriptions "You may struggle to hold a phone number while dialing" - specific to your profile
Actionable recommendations ~ Vague suggestions Specific professionals to see, accommodations to request, strategies to try
Shareable with clinician Chat transcript - not a clinical document Formatted PDF with methodology, disclaimers, and citations
Privacy Your health data sent to OpenAI/Google servers Generated on dedicated infrastructure - your data stays private
Cost ~ 20 $ One-time 14,99 € per report
Pricing

Free assessment, paid reports

Take the tests and see your scores at no cost. If you want the full AI-generated interpretation with pattern analysis, citations, and recommendations, that's the report.

Free Assessment

Free forever

Try select tests from each cognitive domain. See where you stand before unlocking the full picture.

  • Select cognitive tests across 7 domains
  • Score dashboard with radar chart
  • Percentile ranks by age
  • Domain-level performance bands
  • Track your progress over time
  • - No interpretation or explanation
  • - No pattern analysis
  • - No recommendations
RECOMMENDED

AI Report

14,99 € per report

Full cognitive profile with interpretation, pattern analysis, screening indicators, and next steps. Every claim cites real research.

  • Full 26-test battery across all domains
  • Executive summary in plain language
  • Standard scores (M=100, SD=15) per test
  • Domain-by-domain clinical interpretation
  • Cross-domain pattern analysis
  • Dyscalculia & ADHD screening indicators
  • Research citations per section
  • Personalized recommendations
  • PDF export - take it to your clinician
  • Private: processed on dedicated infrastructure
Start Free Tests →
Architecture

Three layers, each doing what it's good at

The system separates math from language on purpose. Numbers are never generated by AI. Prose is never generated without research context. Every layer has a specific job.

1

Deterministic scoring layer

All numbers in your report come from math, not from a language model. Raw scores are converted to standard scores (M=100, SD=15) and percentiles using published normative tables, stratified by age and gender. The AI never computes a number. Sources include Wechsler (WAIS-V), Tombaugh (Trail Making), Strauss, Sherman & Spreen, and the NEURONORMA project for Spanish populations.

2

Retrieval-augmented generation (RAG)

Before writing each section of your report, the engine searches a curated corpus of 220 research papers (468 indexed passages) to find interpretation guidelines, normative context, and clinical precedent relevant to your specific scores. The model writes with the research in front of it, not from memory.

3

Domain-tuned language model

Report prose is generated by a model fine-tuned for neuropsychological interpretation. It writes section by section, each with its own research context, so that every paragraph has relevant citations and every claim traces back to published literature. The model is hosted on dedicated hardware; your data never touches third-party AI services.

4

Cross-domain pattern detection

This is the part that makes the report more than a score-by-score summary. The engine compares your results across all seven cognitive domains to find clinically meaningful patterns. Low numerical cognition with intact reasoning suggests a different profile than low scores across the board. The system uses the NumberSenseMMR framework to distinguish dyscalculia from general math difficulty.

5

Validation and safety checks

Before a report reaches you, automated checks verify that scores are consistent, citations are real, no diagnostic language slipped through, and disclaimers are present. Validity indicators flag potentially unreliable results caused by distraction, misunderstanding, or testing conditions.

Sample output

What the engine produces

Synthetic profile: 34-year-old male, below-average working memory, average reasoning, above-average processing speed. Real scores, real normative comparison, real pattern detection.

neuropsych.io/report/sample
Digit Span (Forward) 77.5
Below Average
Digit Span (Backward) 76.4
Below Average
Stroop (Color-Word) 108.3
Average
Trail Making B 74.1
Below Average
DSST / Coding 124.2
Above Average
Matrix Reasoning 95.0
Average
Verbal Fluency 105.2
Average
Corsi Blocks 80.5
Below Average

From the Pattern Analysis section:

"The profile reveals a distinct dissociation between processing speed and working memory capacity. While the 'engine' runs fast - DSST at the 95th percentile - the ability to hold and manipulate information is significantly impaired. This creates a frustrating paradox where the individual appears quick-witted but fails to retain the necessary details to execute complex plans."

Reference: Bulthé et al. (2019). Multi-method brain imaging reveals impaired representations of number as well as altered connectivity in adults with dyscalculia.

Try it yourself

The cognitive assessment is free. Take the tests, explore your scores, and see if the report adds value for you. 14,99 € per report, no subscription.

Start the assessment →