Research · CEBA engine
Evidence-Based AI for Psychiatry
Evidence-based psychiatry has a delivery problem. Guidelines exist. Trials are indexed. Meta-analyses are one search away. Yet community care often runs off-guideline. CEBA, the Clinical Expert-Based Assistant research engine at Sultan Lab, is how we are building AI that can carry evidence into the prescribing moment instead of producing fluent text that sounds clinical and isn't.
What evidence-based AI requires
1. A curated psychiatry corpus
A defined, versioned, continuously updated body of psychiatric literature: peer-reviewed trials, meta-analyses, society guidelines, and FDA labels. Every paper tagged with design, population, intervention, and outcome metadata.
2. Section-level retrieval
Clinical reasoning depends on structure. A trial in preschoolers is not interchangeable with one in adolescents. Retrieval matches at the section level: population to population, intervention to intervention, outcome to outcome, timing to timing.
3. Study-quality weighting
RCTs and systematic reviews rank above cohort studies and case series. Recency and effect-size adjustments prevent older, highly cited studies from out-voting newer evidence.
4. Traceable citations
Every research output terminates in a real PMID, DOI, or guideline section, so a clinician can read the underlying paper, not just the AI's summary.
CEBA architecture
CEBA-ADHD is the index implementation, combining a graph-based retrieval-augmented knowledge base, expert-committee curation, and patient-specific data integration. Each paper is decomposed into five section-level vectors: overview, population, intervention, outcomes, and timing. A patient case is encoded across the same dimensions, and retrieval is the joint clinical match.
- Generation 1 corpus: approximately 3,000 ADHD papers, indexed and continuously updated
- Weekly PubMed ingestion; ClinicalTrials.gov registrations flagged for in-progress studies
- Society guidelines versioned with effective and superseded dates
- Architecture is condition-agnostic. ADHD is the index condition because its evidence base is exceptionally well-developed
Sigmund's clinical assistant capabilities (Lens, Intake, Note/Scribe, and Scales & Journals) are being built on this research foundation. The evidence engine itself remains under active development at Sultan Lab.
Sigmund is investigational and intended to assist, not replace, clinical judgment. Research described on this page may not yet be available in the product.
Back to research overview