Working today
Eight invented cases, editable finding statuses, five-condition matching, source excerpts, local review history and downloadable reports.
CLINICAL INSIGHT, MADE TRACEABLE
From scattered findings
to a clearer clinical question.
A review workspace for synthetic cases, traceable findings and versioned evidence reports.
05conditions in scope23recognised findings
FROM DEMONSTRATION TO EVALUATION
A working foundation. A defined next step.
Follow a synthetic record through extraction, evidence matching, human review and documented follow-up. The purpose is to demonstrate a reviewable workflow, not diagnostic performance.
Demonstration boundaries
Rule-based matching. No live hospital integration, trained clinical model or measured diagnostic accuracy. Percentages describe mapped findings, not disease probability.
What funding would enable
Clinical knowledge review, an approved evaluation dataset, secure integration engineering and a prespecified independent evaluation. Partner involvement and approvals remain to be agreed.
Suggested demonstration: choose a case → extract and review → match → inspect evidence → record a simulated decision → download the report. Also show the insufficient-evidence case.
LEAD DEMONSTRATION / IMMUNOLOGY REFERRAL SUPPORT
Common variable immunodeficiency (CVID)
Could connecting infection history and antibody findings across visits help clinicians identify who warrants specialist review? This is the question to evaluate, not a proven benefit.
The lead case contains four invented visit entries. The engine matches phrases across the note; it does not perform temporal reasoning, interpret numeric lab results or exclude secondary causes. Age, repeat results, medication effects, protein loss and alternative immune disorders require specialist assessment.
CVID consensus reference | Qatar cohort reference
The Qatar publication includes CVID in a selected malignancy cohort; it does not establish population-wide diagnostic delay. No institutional endorsement. Research rules await clinical review. Not RAG.
SECONDARY DEMONSTRATION / HOMOCYSTINURIA
Classical homocystinuria (CBS deficiency)
Explore two invented cases using a small, source-linked finding map. This is a rule-based addition, not RAG, and it has not been clinically reviewed or validated.
Homocystinuria is documented in Qatar and is already part of newborn screening. This demonstration does not replace screening or show that screening failed. A metabolic specialist must define the additional clinical use case.
Clinical reference: GeneReviews · Qatar context and screening: HMC
Sources checked 6 October 2026. References are not institutional partnerships or endorsements. No treatment recommendations or diagnostic probabilities are generated.
Clinical workflow demonstration · October 2026. All stages use synthetic data. Data transfer, clinical actions and learning are simulated; no hospital connection or automatic model training.
Case selectedFindings extractedEvidence matched
01 · Patient encounter
Local workspace02 · Data capture
Manually entered or preloaded synthetic text. No EHR, FHIR or HL7 connection.
Try changing a finding to “No tremor” or “Possible proteinuria” to see how context changes the output.
Demo scope & limitations
This is a rule-based prototype, not a trained clinical AI model. It recognises a small English vocabulary and simple negation, uncertainty and family-history phrases. It does not understand full clinical context, temporal relationships, lab ranges, genetics or other diseases.
Review all extracted findings before matching. Missing information is never treated as an absent finding.
03 · RareAI analysis
Awaiting extractionConfirm or correct the status. Family history and uncertain findings do not count as patient matches.
Select a case and extract findings to begin.
EVIDENCE EXPLORER
04 · Results for clinician
Matches are limited to five selected conditions. Percentages show finding coverage, not disease likelihood, risk or diagnostic confidence. Ranking uses present finding counts.
Reviewed findings will appear here as traceable condition matches.
Evidence trail
Excerpts come from the current synthetic narrative. Disease references support the general knowledge mapping; they are not evidence that this patient has a disease.
05 · Clinical review & decision
Record the clinician's simulated decision independently of the matching output. No orders or referrals are sent.
06 · Follow-up & learning
Feedback is saved for human review only. It does not retrain or change the matching engine. A pending follow-up can be updated in a later saved version.
Case history
Built for a clinical conversation
A clinician defines the use case and reference diagnoses; Sefar supplies the engineering workflow. This MVP does not assert a confirmed institutional partnership.
Next validation gate
Clinical review of the knowledge set, approved data access, independently labelled records and a prespecified evaluation protocol before any patient-facing testing.