Research Initiative · Early Stage MVP

Auditing bias in how we diagnose children

ATLAS is a diagnostic equity auditing platform that flags racial, gender, and cultural bias in neurodevelopmental assessments — before they shape a child's educational trajectory.

Collaborate with Us See How It Works

The Problem

The tools clinicians use weren't built for every child

Gold standard neurodevelopmental assessments like Vanderbilt and ADOS screenings used in schools nationwide carry documented demographic bias. Yet no system exists to audit this at the point of evaluation.

35%
more often ODD is diagnosed in Black children than white children. This is despite similar actual prevalence rates across racial groups. The same behaviors read as ADHD in white children and conduct disorder in Black children.
Williams et al. (2024) — analysis of 849,000+ patients across 50 U.S. health systems
1.8 yrs
later on average that autistic girls are diagnosed compared to boys. Girls also seem to require a more severe clinical presentation to receive a diagnosis at all.
Begeer et al. (2013), J. Autism Dev. Disord.
Zero
dedicated auditing tools exist at the point of evaluation to translate bias research into real-time clinical decisions.
ATLAS project thesis

How It Works

From literature to logic

ATLAS flags directly reference peer-reviewed literature. Here's how published findings become real-time flags inside the engine.

01 — LITERATURE
Research identified
Peer-reviewed DIF studies, cross-cultural validity research, and informant bias literature are reviewed and mapped to instruments used by districts. We're actively building this list and looking for practitioners who can tell us what's actually in use.
02 — EXTRACTION
Bias patterns codified
Key findings are extracted into structured flags: which subscales show differential item functioning, which populations are affected, and under what conditions scores are unreliable.
03 — HEURISTIC LOGIC
Rules built into the engine
Each flag becomes a conditional rule. When a child's profile matches a known bias pattern, ATLAS surfaces the relevant finding with the source citations.
04 — VALIDATION
Tested against real cases
Heuristic rules are only as good as the data they're tested on. Each flag needs to be verified against real evaluation records to confirm accuracy. This is the step we're actively seeking research partners for.

Where We Are

Early stage and looking for partners.

✓
Literature review complete (ADHD) Core DIF studies, cross-cultural validity papers, and informant bias research mapped to equity flags.
DONE
✓
Working MVP built Heuristic logic engine flags known bias patterns for ADHD assessments. Vanderbilt ADHD Rating Scale fully implemented.
DONE
◉
Expanding to autism screeners Adding GARS, ADOS supplementary heuristics, and gender bias flags. Known gap: ELL/bilingual validity literature is thin.
IN PROGRESS
◉
Seeking faculty mentorship / advisor Looking for a researcher in school psych, special ed, or psychometrics to guide the validation methodology.
SEEKING

What we need right now

  • School psychologists, evaluators, special ed directors, someone who can tell us which assessment instruments their district actually uses
  • Access to de-identified evaluation records to validate the heuristic logic against real cases

Collaborate

Who we're looking for

Reach out with your expertise in the field:

🎓
Faculty Researchers
School psychology, special education, psychometrics, health equity. Your expertise is valued, whether you want to give quick feedback or to be a formal mentor.
WHAT WE'RE ASKING → Review the prototype. Poke holes in the methodology. Help with the research process.
🏫
School Psychologists & Evaluators
Practicing clinicians who run neurodevelopmental evaluations in K-12 settings. Your floor-level knowledge of what actually happens in evaluations is what we're missing most.
WHAT WE'RE ASKING → Walk through one case with us. Give your inputs on the diagnostic process.
🏛️
Special Education Directors
District-level leaders managing compliance, IEP teams, and disproportionality requirements. Especially relevant for California LEAs currently under CDE monitoring.
WHAT WE'RE ASKING → A 30-minute conversation about your disproportionality workflow and whether ATLAS could fit in it.

This is the moment to shape it

ATLAS is early enough that your input could define the methodology. Make the difference and get in touch.

Request a Demo or Conversation talha.mansari808@gmail.com sy.umair.ahmed@gmail.com