The challenge
Patient demand for Adult ADHD evaluations has surged in recent years, and the clinical preparation process behind them has not kept pace. For MindCare Health, providers were spending 2-3 hours per patient just preparing for initial consultations.
Before each appointment, providers had to manually review lengthy intake forms, score standardized assessments by hand, compile clinical notes, and prepare patient-friendly summaries. For a clinic processing 20-30 evaluations per week, this administrative burden was preventing providers from doing what they do best: spending time with patients.
The manual process was also inconsistent. Different providers focused on different parts of the same intake form, so how thoroughly a case was prepared depended on how much time that provider had.
MindCare needed the pre-assessment work streamlined so providers could focus on clinical judgment and patient care.
The solution
We built an automation system that handles the administrative preparation work, giving providers a head start before each patient appointment.
When a patient submits their intake form through MindCare's patient portal, the system automatically processes the submission, scores standardized assessments, and organizes the information into structured reports.
Providers receive a summary document that highlights key information from the intake form, presents assessment scores in a clear format, and suggests areas to explore during the clinical interview. A separate post-appointment document explains, in plain language for the patient, what the assessment covers and what to expect.
The system handles the time-consuming data organization so providers can spend their preparation time on clinical thinking rather than paperwork. Every clinical decision stays with the provider: the system organizes the inputs and makes no diagnostic determination.
AI architecture
The first finding on this build was how little of it needed a model at all. Roughly eighty percent of the report is scoring: the ASRS and the executive function items are validated instruments with published, deterministic scoring rules, and arithmetic does that better than inference. So the scoring runs in code, and the model handles the one part that genuinely benefits from language: organizing the findings into something a provider can read in a few minutes.
Rather than generic summaries, the model references specific details from each patient's intake form, so a report reflects the person in front of the provider rather than a template.
On privacy, the payload sent for processing is a limited data set rather than full de-identification. Direct identifiers are stripped before anything leaves the system. Free-text narrative fields can still carry a third-party name, an employer, a school, or a date a patient chose to write, and that is documented as a known limit rather than a solved problem. Data handling, audit logging, and retention meet HIPAA requirements.
If the model is unavailable, a rule-based scorer takes over. It makes no network calls, and it stamps its own output to say simplified scoring was used, because a degraded report that looks identical to a normal one is worse than a delay.
The results
Work that took a provider 2+ hours of manual compilation now completes in about 5 minutes of automated processing, and what is left for the provider is reviewing a finished document rather than assembling one. The standardized format means they can scan a report quickly and spend their attention on the parts that need clinical judgment.
At roughly $0.08 per evaluation, the processing cost is minimal compared to the staff time it saves. The clinic handles 20-30 evaluations weekly without the administrative bottleneck they had before.
Automated audit logging and data retention meet HIPAA compliance requirements, and the system scales automatically during high-volume periods.