The AI report generation system is the core of the platform. Once an assessment’s sections have been completed, the platform processes all answers through a structured pipeline and produces a comprehensive clinical narrative document — ready for clinician review and payer submission.
How It Works
Report generation is triggered by the clinician from within the assessment editor. The system runs as a background process, allowing the user to monitor progress without waiting at a loading screen.
The generation pipeline follows these steps:
Step 1 — Answer Collection
The system retrieves all answers saved across every section of the assessment, including answers from repeatable sections (each form of maladaptive behaviors, each replacement program).
Step 2 — Subject Grouping
Answers are grouped by clinical subject area. Each subject corresponds to a domain of the behavioral evaluation — sleep and health, communication, daily living skills, social interaction, maladaptive behaviors, replacement programs, and others.
There are twenty-four distinct subject areas, each processed independently. This ensures that generated narrative sections remain focused and clinically coherent rather than blending unrelated information.
Step 3 — AI Narrative Generation
For each subject, the system sends a structured prompt to OpenAI GPT-4 containing:
- The questions and answers for that subject
- Any clinician-entered explanations
- Clinical context specific to that domain
GPT-4 generates a professional clinical narrative for each subject — written in the appropriate tone and style for insurance and payer documentation.
Repeatable subjects (maladaptive behaviors and replacement programs) generate a separate narrative block for each form instance. A client with three maladaptive behaviors receives three independently written behavior analysis sections.
Step 4 — Progress Broadcasting
As each subject is processed, a real-time progress update is sent to the clinician’s browser via WebSocket. A progress bar advances as sections complete, so the clinician always knows how far along generation is.
Step 5 — Document Assembly
Once all subject narratives are generated, the platform assembles the complete report:
- Clinical narratives are placed into the appropriate sections of the document template
- Baseline graphs are embedded as images
- Uploaded files and logos are positioned in the document
- ICD-10-CM codes are placed in the diagnostic section
- The document is formatted for professional presentation
Step 6 — Document Delivery
The finished Word (.docx) report is:
- Stored securely in AWS S3
- Made available via a time-limited, signed CloudFront download link
- Sent to the clinician’s registered email address as a notification
The case status updates to Completed and the download button becomes active.
Caching for Efficiency
AI-generated responses are cached for seven days. If a report is regenerated — for example, after minor answer adjustments — the system reuses cached responses for subjects that have not changed, reducing generation time and API cost.
Draft vs. Final Reports
Clinicians can request a preview before generating the final report. The preview generates draft narratives for review without marking the case as complete, giving the clinician the opportunity to:
- Review AI-drafted content
- Return to the assessment and adjust answers if needed
- Add or modify clinical notes in editable fields
- Then generate the final report when satisfied
AI-Assisted Answer Review
For rich text editor fields, an optional AI review feature is available on individual answers. When enabled, the clinician can submit a written response and receive AI feedback suggesting improvements for clarity, clinical tone, or completeness. This is a per-answer feature, separate from full report generation.
What the AI Generates
The AI drafts the following components of the report:
| Component | Description |
|---|---|
| Behavior analysis narratives | Clinical descriptions of each maladaptive behavior, including topography, function, and intensity |
| Replacement program descriptions | Narrative descriptions of each replacement behavior goal and its rationale |
| Domain summaries | Synthesized summaries for each clinical subject area (sleep, communication, daily living, etc.) |
| Intervention recommendations | Evidence-based intervention strategies appropriate to the presenting behaviors |
| Clinical impressions | Summary narrative integrating findings across the assessment |
All generated content is editable in the final Word document. The clinician retains full control over the report and is responsible for reviewing and approving all AI-drafted content before submission.
Important Notes
- The AI does not have access to external client records or prior assessments
- All generated content is based solely on the answers entered in the current assessment
- No personally identifiable client information is sent to OpenAI — prompts use de-identified clinical data
- The platform does not guarantee clinical accuracy; all generated content must be reviewed by a qualified BCBA before use
