PsychNET prototype · synthetic data
Not sure if something is wrong? Ask. Ready to act? Screen. Either way, PsychNET gets your child to the right provider faster.
Describe what you're seeing in plain language. PsychNET will tell you whether it's typical development or worth investigating — grounded in CDC milestones and DSM-5 norms, not generic parenting advice. When the conversation crosses a clinical threshold, one tap starts a formal screening with your context already filled in.
5-step caregiver flow. The orchestrator agent picks which validated instruments to use based on the intake concern, then asks one question at a time — stopping early when the diagnosis is decided. Ends with triage, a provider match, and a full psychiatric-eval note ready to hand to the clinician.
What the system has noticed across all closed journeys: which questions yielded the most information, which forms picked the right instrument first, which orchestrator routes were correct. Useful for researchers and ops. Read-only.
Explore the data →For any presenting concern, see how many items the full battery would ask vs. how many PsychNET's adaptive engine actually asks — and whether the conclusion changes. Honest equivalence metric (mean imputation, not zero-extension).
Run a comparison →What a clinician actually receives. Psychiatric-evaluation format with HPI, screening results (item-level expandable), assessment, recommendations, and a full agent decision log so the AI shows its work.
View a sample note →Specialty agents (child psychiatry active; pediatrics, neurology, cardiology, GI in preview) and the orchestrator that routes between them.
Per-question information value across the corpus — the lever that decides which items the adaptive engine asks first.
How the system picks which specialty agent should handle a referral, and how often the rules engine and Claude agree.